Optical packet buffering device and method

Optical packet buffering device and method
US7693421

A device that accepts input of asynchronously-arriving variable-length optical packets transmitted over a plurality of lightpaths, and outputs same to a single optical path, comprising: in order to prevent the optical packets from overlapping in the output lightpath, a controller that uses the delay times of the delay elements and the optical packet length and arrival gap time thus read to determine by computation the delay element used for temporary storage, where a plurality of stages of processors is provided. Thus, the processing required to determine the delay time is performed by parallel processing with the results of processing the prefix-sum operation used in parallel pipelined processing along with the queue length, optical packet length and arrival gap time of the buffering device.

PTO Wrapper PDF
Dossier Espace Google

Patent 7693421
Priority Jul 09 2003
Filed Jul 09 2004
Issued Apr 06 2010
Expiry Feb 05 2027 Extension 941 days
Inventors Harai, Hir…
Assg.orig National I…
Assg.curr National I…
Entity Large
Referenced by 2
References 1
Maint.: EXPIRED

BACKGROUND OF THE IN…
SUMMARY OF THE INVEN…
BRIEF EXPLANATION OF…
DESCRIPTION OF THE P…
Example 1
Example 2
Example 3
Example 4

13. An optical packet buffering device that accepts input of synchronous fixed-length optical packets from N light paths, wherein N denotes an integer of 2 or more, and gives each of the optical packets one of B different delay times, wherein B denotes an integer of 1 or more, and outputs the synchronous fixed-length optical packets from N light paths to a single light path, comprising:

means for reading a presence of optical packets on the respective light paths;

a queue consisting of a plurality of delay elements that temporarily store said plurality of optical packets and give each optical packet one of the B different delay times;

an optical switch that directs each of the optical packets to one of said plurality of delay elements;

an optical multiplexer connected by light paths to said plurality of delay elements;

an output light path connected to the multiplexer;

a controller which uses the delay times of the delay elements and the length of the optical packets and the presence information for optical packets read by the means for reading to select the delay elements used for temporary storage from said plurality of delay elements in order to prevent the accepted optical packets from overlapping in the output light path and gives respectively sequential numbers n from 1 to N to said N light paths, wherein,

in said controller, a length l_nand an arrival gap time t_n, larger than or equal to 0, of the optical packets of the nth light path that pass through the means for reading at a specified time are respectively read at each period t, both l_nand t_nare set to be zero when the optical packets of the n^thlight path do not arrive at the means for reading during an initial period t, and, with respect to n−1=0, assume l_n−1=t_n−1=0;

an arrangement of a first stage, in which processors p(1,n) that each perform a calculation using the optical packet presence information on the light path of the optical packet length l_n−1and the arrival gap time t_n−1and the optical packet presence information on the light path of the optical packet length l_nand the arrival gap t_n, and registers R(1,n) store the results of calculation;

an arrangement of an m^th(2≦m≦log₂N) stage, which has processors p(m,n) that calculate the contents of the registers R(m−1,n−2^m−1) or copies of the registers R(m−1,n2^m−1) and the registers R(m−1,n) or copies of the registers R(m−1,n), and registers R(m,n) that store the results of the calculation;

a register q that stores information at a tail end of a queue length which is updated when the processors perform operations of adding all arriving optical packets to the queue;

a processor p(log₂N+1,1) that determines the delay applied to the first synchronous fixed-length optical packet by performing calculation on the basis of the optical packet presence information on the first light path or a copy of the optical packet presence information on the first light path and the value of the register q;

processors p(log₂N+1,n), for values of n greater than 1, that determine the delay applied to the n^thsynchronous fixed-length optical packet by performing calculation on the basis of the optical packet presence information on the n^thlight path or a copy of the optical packet presence information on the n^thlight path and the values of the registers R(log₂N,n−1 and the register q and;

an arrangement of a (log₂N+1)^thstage, in which delay times applied to the synchronous fixed-length optical packets of the respective light paths are coordinated based on the output of the processor p(log₂N+1,1) or the processors p(log₂N+1,n).

18. An optical packet buffering device that accepts input of synchronous fixed-length optical packets from N light paths, wherein N denotes an integer of 2 or more, and gives each of the optical packets one of B different delay times, wherein B denotes an integer of 1 or more, and outputs the packets to a single light path, comprising:

means for reading a presence of optical packets and priority-class information of said optical packets on the respective light paths;

a queue consisting of a plurality of delay elements that temporarily store said plurality of optical packets and give each optical packet one of the B different delay times;

an optical switch that directs each of the optical packets to one of said plurality of delay elements;

an optical multiplexer connected by light paths to said plurality of delay elements;

a controller which uses the delay times of the delay elements, the presence information of the optical packets and the priority-class information of said optical packets read by the means for reading to select the delay element used for temporary storage from said plurality of delay elements in order to prevent the accepted optical packets from overlapping in an output light path and gives respectively sequential numbers n from 1 to N to said N light paths, wherein

in said controller, a length l_nand an arrival gap time t_n, which is larger than or equal to 0, of the optical packets of the n^thlight path that pass through the means for reading at a specified time are respectively read at each period t, both l_nand t_nare set to be zero when the optical packets of the n^thlight path do not arrive at the means for reading during an initial period t, and, with respect to n−1=0, assume l_n−1=t_n−1=0;

an arrangement of a first stage, in which processors p(1,n) each perform a calculation using the optical packet presence information and priority-class information on the light path of the length l_n−1and the arrival gap time t_n−1 and the optical packet presence information and priority-class information on the light path of the length l_nand the arrival gap time t_n, and registers R(1,n) store the results of calculation;

an arrangement of an m^th(2≦m≦log₂N) stage, which has processors p(m,n) that calculate the contents of the registers R(m−1,n−2^m−1) or copies of the registers R(m−1,n−2^m−l) and the registers R(m−1,n) or copies of the registers R(m−1,n), and registers R(m,n) that store the results of the calculation;

a register q that stores information at the tail end of the queue length which is updated when the processors perform operations of adding all arriving optical packets to the queue;

a processor p(log₂N+1,1) that determines the delay applied to the first synchronous fixed-length optical packet by performing calculation on the basis of the optical packet presence information and the optical packet priority-class information on the first light path or a copy of the optical packet priority-class information on the first light path and the value of the register q;

processors p(log₂N+1,n), for values of n greater than 1, that determine the delay applied to the n^thsynchronous fixed-length optical packet by performing calculation on the basis of the optical packet presence information and optical packet priority-class information on the n^thlight path or a copy of the optical packet priority-class information on the n^thlight path and the values of the registers R(log₂N,n−1) and the register q; and

an arrangement of a (log₂N+1)^thstage, in which delay times applied to the synchronous fixed-length optical packets of the respective light paths are adjusted based on the output of the processor p(log₂N+1,1) or the processors p(log₂N+1,n).

1. An optical packet buffering device that accepts input of asynchronously-arriving variable-length optical packets transmitted via N light paths, wherein N denotes an integer of 2 or more, and outputs the asynchronously-arriving variable-length optical packets to a single light path, comprising:

means for reading optical packet lengths of the variable-length optical packets on the individual light paths;

means for reading an arrival gap time between a start time of a clock signal prepared in advance and an arrival time of the subsequent most recent variable-length optical packet;

a queue consisting of a plurality of delay elements that temporarily store said plurality of optical packets and give each of said optical packets one of B different delay times, wherein B denotes an integer of 2 or more;

an optical switch that allots said optical packets individually to one of said plurality of delay elements;

an optical multiplexer that is connected by light paths to said plurality of delay elements and enabled to multiplex the optical packet signals delayed by the delay elements;

an output light path connected to the multiplexer;

a controller which, by using the delay times of the delay elements and the lengths of the optical packets and the arrival gap time read by the means for reading the arrival gap time, selects the delay element used for temporary storage from said plurality of delay elements in order to prevent the accepted optical packets from overlapping in the output light path, wherein

the controller gives sequential numbers n from 1 to N individually to said N light paths, and

the means for reading the optical packet lengths of variable-length optical packets and the means for reading the arrival gap time respectively read with a period t, a length l_nof an optical packet and the arrival gap time t_n, which is larger than or equal to 0, of the optical packets of an n^thlight path that passes through the means for reading the optical packet lengths at a specified time and, with respect to the optical packets that fail to reach the means for reading the optical packet lengths during an initial period t, set both l_nand t_nto be zero, and, with respect to n−1=0, assume l_n−1=t_n−1=0;

an arrangement of a first stage, which has disposed therein processors p(1,n) that calculate the optical packet length l_n−1and the arrival gap time t_n−1and the optical packet length l_nand the arrival gap time t_n, and registers R(1,n) that store the results of calculation;

an arrangement of an m^th(2≦m≦log₂N) stage, which has disposed therein processors p(m,n) that calculate the contents of the registers R(m−1,n−2^m−1) or copies of the registers R(m−1,n−2^m−1) and the registers R(m−1,n) or copies of the registers R(m−1,n), and registers R(m,n) that store the results of the calculation; and

an arrangement of a (log₂N+1)^thstage, which is provided with a register q that stores the information at a tail end of a queue length which is updated when the processors perform operations of adding all arriving optical packets to the queue, also, a processor p(log₂N+1,1) that determines the delay applied to the first variable-length optical packet by performing calculation on the basis of the optical packet length and arrival gap time for the first light path or a copy of the arrival gap time for the first light path and a value of the register q, and processors p(log₂N+1,n) that determine the delay applied to the n^thvariable-length optical packet by performing calculation on the basis of the optical packet length l_nand arrival gap time t_nfor the n^thlight path or a copy of the arrival gap time t_nfor the n^thlight path and the values of the registers R(log₂N,n−1)and the register q, wherein

in the arrangement of the (log₂N+1)^thstage, the delay times applied to the variable-length optical packets of the individual light paths are adjusted based on the output of the processor p(log₂N+1,1) or the processors p(log₂N+1,n).

2. The optical packet buffering device according to claim 1, wherein

the means for reading the optical packet length of variable-length optical packets performs reading of the length l_nat each period t and the means for reading the arrival gap time performs reading the arrival gap time t_n, which is larger than or equal to 0, of the optical packets of the n^thlight path that pass through the means for reading the optical packet lengths at a specified time and both l_nand t_nare set to zero when the optical packets of the nth light path do not arrive at the means for reading the optical packet lengths during an initial period t, and

with respect to n−1=0, assume l_n−1=t_n−1=0,

the plurality of said delay elements give each of the optical packets delays that respectively increase by a delay time D>0;

each of said processors is provided on one of the respective N light paths on the input side, and perform the following processing on the integers n from 1 to N:

1) when l_n−1=0 is true, output t_nas t_n,1, or

when l_n−1=0 is not true, output t_n−1as t_n,1,

2) and take g_n,1=1_n−1+t_n−1,

when l_n=0 is true, and output g_n−1as last position f_n−1of the optical packets of the n^thlight path, or

when l_n=0 is not true, taking Δ_nto be the smallest integer no less than (g_n−1−t_n)/D set (t_n+l_n+Δ_nD) to the output f_n,1of the last position of the optical packets of the n^thlight path.

3. The optical packet buffering device according to claim 2, wherein

under a condition that N is an integer of 3 or more,

each of the processors p(1,n) in the arrangement of the first stage is respectively provided with a register R(1,n) that temporarily stores its respective output of (f_n,1, t_n,1), and

the processors p(2, n) in the arrangement of the second stage, for the integers n from 3 to N, calculate the content of the n^thregister and the content of the (n−2)^thregister and gives the output of (f_n,2, t_n,2), where

1) when f_n−2,1=0 is true, output t_n,1as t_n,2, or

when f_n−2,1=0 is not true, output t_n−2,1as t_n,2,

2) in addition, when f_n,1=0 is true, output f_n−2,1as f_n,2, or

when f_n,1=0 is not true,

2-1) taking Δ_nto be the smallest integer no less than (f_n−2,1−t_n,1)/D set

(f_n,1+Δ_nD) to the output f_n,2.

4. The optical packet buffering device according to claim 2, wherein

each of the processors p(1,n) in the arrangement of the first stage is respectively provided with a register R(1,n) that temporarily stores its respective output of (f_n,1, t_n,1), and

the processors p(m, n) in the arrangement of the (m−1)^thstage, for the integers m from 2 to log₂N, are provided,

for the m^thstage processors, and for the integers n from β+1 to N where β=2^m−1, the (n−β)^thprocessors p(m,n) in the arrangement of the m^thstage calculate the content of said n^thregister R(m−1,n) connected to the previous-stage processors p(m−1,n) and the content of the (n−β)^thregister R(m−1,n−β) and gives the output of (f_n,m, t_n,m), where

1) when f_n−β,m−1=0 is true, output t_n,m−1as t_n,m, or

when f_n−β,m−1=0 is not true, output t_n−β,m−1as t_n,m,

2) in addition, when f_n,m−1=0 is true, output f_n−β,m−1as f_n,m, or

when f_n,m−1=0 is not true,

2-1) taking Δ_nto be the smallest integer no less than (f_n−β,m−1−t_n,m−1)/D set

(f_n,m−1+Δ_nD) to the output f_n,m.

5. The optical packet buffering device according to claim 4, wherein

M=log₂N and the length of the queue is defined as q, for the integers n from 1 to N,

1) when no optical packet exists, or l_n=0 is true, end processing, and

when no optical packet exists, for f_n−1,Min the register R(M,n−1),

2) when f_n−1,M=0 is true, provide a temporary value (temp), and set (temp)=q,

when f_n−1,M=0 is not true, set the sum of f_n−1,Mand the smallest integer no less than (q−t_n,−1,M)/D to (temp),

3) taking Δ_nto be the smallest integer no less than ((temp)−t_n)/D, when Δ_nis less than B, output Δ_nas information that specifies a delay element, but if not less, discard the corresponding optical packet.

6. The optical packet buffering device according to claim 5, wherein

1) when f_N,M=0 is true for the f_N,Mstored in register R(M,N), the larger of (q−T) or 0 is stored as the new queue length (q_new) in the register q,

when f_N,M=0 is not true, Δ is set as the smallest integer no less than (q−t_N,M)/D and at this time,

2) when this Δ is smaller than B, store the value of the larger of (f_N,M+ΔD−T) and 0 in the register q as the new queue length,

when this Δ is not smaller than B, take (q−T) as the new queue length and store it in the register q.

7. The optical packet buffering device according to claim 5, wherein

an MTU (maximum transfer unit) is set in optical packet communications,

1) when f_N,M=0 is true for the f_N,Mstored in register R(M,N), the larger of (q−T) or 0 is stored as the new queue length (q_new) in the register q,

when f_N,M=0 is not true, Δ is set as the smallest integer no less than (q−t_N,M)/D and at this time,

2) when this Δ is smaller than said B, store the value of the larger of 0 and the value found by subtracting t from the lesser of (f_N,M+ΔD) and (BD+MTU) in the register q,

when this Δ is not smaller than B, take (q−T) as the new queue length and store it in the register q.

8. The buffering device for asynchronously-arriving optical packets according to claim 1, wherein

and the period t is set as a power of 2 times a set time unit (τ) set in advance, or namely T=2^aτ (where a is a non-negative integer), while t is to be no greater than the minimum packet length represented as the length of an integral multiple of the time unit (τ),

the means for reading the optical packet length of variable-length optical packets, read with a period t, perform reading of the length l_nand the means for reading the arrival gap time, read with a period t, perform reading the arrival gap time t_nof the optical packet in the n^thlight path that pass through the means for reading the optical packet lengths at a specified time, the length l_nis represented as a length which is an integral multiple of the time unit (τ) that is not less than the length of the optical packet in the n^thlight path, the arrival gap time t_nis a length which is an integer multiple of the time unit (τ) that is not more than the arrival gap time of the optical packet in the n^tlight path, and for n−1=0, l_n−1=t_n−1=0,

the plurality of said delay elements give each of the optical packets delays that respectively increase by the delay time D which is given as 2^kτ (where k is a non-negative integer), and, the integer corresponding to the k least significant bits of a binary representation of an integer x is represented as x⁽²⁾, and a binary number consisting of the remaining most significant bits after removing the aforementioned k least significant bits is represented as x⁽¹⁾, each of said processors is provided on one of respective N light paths on the input side, and performs the following processing for the integers n from 1 to N:

1) when l_n−1=0 is true, output t_nas t_n,1, or

when l_n−132 0 is not true, output t_n−1as t_n,1,

2) and take g_n,1=l_n−1+t_n,1,

when l_n=0 is true, output g_n−1as f_n,1, or

when l_n=0 is not true,

i) when g_n−1is less than t_n, set g_n−1⁽¹⁾to Δ_n,

ii) when g_n−1is not less than t_n, set (g_n−1⁽¹⁾+1) to Δ_n, and

3) set (t_n⁽²⁾+l_n+Δ_nD) to the output f_n,1.

9. The optical packet buffering device according to claim 8, wherein

each of the processors p(1,n) in the arrangement of the first stage is respectively provided with a register R(1,n) that temporarily stores its respective output of (f_n,1, t_n,1);

processors are provided, for the integers m from 2 to log₂N, in the arrangement of the (m−1)^thstage;

for the m^th-stage processors, and for the integers n from β+1 to N where β=2^m−1, the (n−β)^thprocessors p(m, n) in the arrangement of the m^th-stage calculate the content of said n^thregister R(m−1,n) connected to the previous-stage processors p(m−1,n) and the content of the (n−β)^thregister R(m−1,n−β) and gives the output of (f_n,m, t_n,m), where

1) when f_n−β,m−1=0 is true, output t_n,m−1as t_n,m, or

when f_n−β,m−1=0 is not true, output t_n−β,m−1as t_n,m,

2) in addition, when f_n,m−1=0 is true, output f_n−β,m−1as f_n,m, or

when f_n,m−1=0 is not true,

2-1) when f_n−β,m−1⁽²⁾is less than or equal to t_n,m−1⁽²⁾, set f_n−β,m−1⁽¹⁾to Δ_n,

2-2) when f_n−β,m−1⁽²⁾is not less than or equal to t_n,m−1⁽²⁾, set (f_n−β,m−1⁽¹⁾+1) to Δ_n,

3) set (f_n,m−1+2^kΔ_n) to the output f_n,m.

10. The optical packet buffering device according to claim 8, wherein

M=log₂N and the length of the queue is defined as q⁽ⁿ⁾, for the integers n from 1 to N, and, the integer corresponding to the k least significant bits of the binary representation of the integer x is represented as x⁽²⁾, and the binary number consisting of the remaining most significant bits after removing the aforementioned k least significant bits is represented as x⁽¹⁾

1) when no optical packet exists, or l_n=0, end processing,

2) when an optical packet exists,

2-1) when f_n−1,M=0 is true and,

when q⁽²⁾is less than or equal to t_n⁽²⁾, set q⁽¹⁾to Δ_n,

when q⁽²⁾is not less than or equal to t_n⁽²⁾, set (q⁽¹⁾+1) to Δ_n,

2-2) when f_n−1,M=0 is not true and,

when q⁽²⁾is less than or equal to t_n−1,M⁽²⁾, set q⁽¹⁾to Δ₁,

when q⁽²⁾is not less than or equal to t_n−1,M⁽²⁾, set (q⁽¹⁾+1 to Δ₁, and moreover,

2-3-1) when f_n−1,M⁽²⁾is less than or equal to t_n⁽²⁾,

set (Δ₁+f_n−1,M⁽¹⁾) to Δ_n,

2-3-2) when f_n−1,M⁽²⁾is not less than or equal to

t_n⁽²⁾, set (Δ₁+1+f_n−1,M⁽¹⁾) to Δ_n,

3) when Δ_nis less than B, output Δ_nas information that specifies a delay element,

but if not less, discard the corresponding optical packet.

11. The optical packet buffering device according to claim 9, wherein

M=log₂N and the length of the queue is defined said q⁽ⁿ⁾, for the integers n from 1 to N,

and, the integer corresponding to the k least significant bits of the binary representation of the integer x is represented as x⁽²⁾, and the binary number consisting of the remaining most significant bits after removing the aforementioned k least significant bits is represented as x⁽¹⁾

1) when f_N,M=0 is true,

1-1) when q is greater than t, (q−T) is set as the new queue length (q_new),

1-2) when q is not greater than t, q_new=0 is set,

2) when f_N,M=0 is not true,

2-1) when q⁽²⁾is less than or equal to t_N,M⁽²⁾, q⁽¹⁾is set as Δ, when q⁽²⁾is not less than or equal to t_N,M⁽²⁾, (q⁽¹⁾+1) is set as Δ,

3-1) when this Δ is smaller than B and,

3-1-1) when f_N,M+2^kΔ is greater than t, (f_N,M+2^kΔ−T) is set as q_new,

3-1-2) when f_N,M+2^kΔ is not greater than t, q_new=0 is set,

3-2) when this Δ is not smaller than B, (q−T) is set as q_new.

12. The optical packet buffering device according to claim 9, wherein

an MTU (maximum transfer unit) is set-in optical packet communications;

M=log₂N and the length of the queue is defined as said q⁽ⁿ⁾, for the integers n from 1 to N;

1) when f_N,M=0 is true,

1-1) when q is greater than t, (q−T) is set as the new queue length (q_new),

1-2) when q is not greater than t, q_new=0 is set,

2) when f_N,M=0 is not true,

2-1) when q⁽²⁾is less than or equal to t_N,M⁽²⁾, q⁽¹⁾is set as Δ,

when q⁽²⁾is not less than or equal to t_N,M⁽²⁾,

(q⁽¹⁾+1) is set as Δ,

3-1) when this Δ is smaller than B and,

3-1-1) when (f_N,M+2^kΔ) is greater than

(2^kB+MTU), (2^kB+MTU−T) is set as q_new,

3-1-2) when (f_N,M+2^kΔ) is not greater than (2^kB+MTU) and,

3-1-2-1) when f_N,M+2^kΔ is greater than

(f_N,M+2^kΔ−T) is set as q_new,

3-1-2-2) when f_N,M+2^kΔ is not greater than t, q_new=0 is set,

3-2) when this Δ is not smaller than B, (q−T) is set as q_new.

14. The buffering device for optical packets according to claim 13, wherein said means for reading, for each period t, reads the presence of optical packets on the respective light paths;

the respective processors of the arrangement of the first stage input variables l_n−1and l_naccording to the presence of optical packets arriving at a port (n−1) corresponding to the (n−1)^thlight path and a port n corresponding to the n^thlight path, wherein each variable is 1 when a packet has arrived and 0 when no packet has arrived, and perform the following processing:

1₀=0

for each processor n, in parallel (n=1 to N) f_n,1=l_n−1+l_n;

and provide output of f_n,1as the number of packets arriving at the two ports.

15. The buffering device for optical packets according to claim 14, wherein,

each of the respective processors p(1,n) of the arrangement of the first stage is provided with a register R(1,n) that temporarily stores output (f_n,l) of the respective processors p(1,n), and

each of the respective processors of the arrangement of the (m−1)^thstage are provided for the integers m from 2 to log₂N;

the processor p(m,n) in the arrangement of the m^th-stage, for the integers n from to β+1 to N where β=2^m−1uses the value of the n^thregister R(m−1,n) connected to the processor p(m−1,n) in the arrangement of the respective previous-stage and the value of the (n−β)^thregister R(m−1,n−β) connected to processor p(m−1,n−β) and performs the processing:

f_n,m=f_n−m−1+f_n−β,m−1and provides output of f_n,m.

16. The buffering device for optical packets according to claim 15, wherein, M=log₂N and

the processors p(M+1,n) (where 1≦n≦N in the arrangement of the the (m+1)^th-stage use

f_n−1,Mas the number of optical packets arriving from ports 1 to n−1,

l_nwhich indicates the presence of an optical packet arriving at port n, and the queue length q;

perform the following processing respectively for the integers n from 1 to N:

if no optical packet exists, or l_n=0, end processing;

else Δ_n=q+f_n−1,M;,

but if Δ_n<B packet n is given delay Δ_n;

and if not Δ_n<B packet n is discarded;

and output the value of the delay Δ_nas information that specifies the delay element.

17. The buffering device for optical packets according to claim 16, wherein, for the f_N,Mstored in register R(M,N), the larger value between (min(q+f_n,M,B)−1) and 0 is stored as the new queue length (q_new) in register q.

19. The buffering device for optical packets according to claim 18, wherein the respective processors of the arrangement of the first stage input variables l_n−1and l_nthat indicate the presence of optical packets arriving at a port (n−1) corresponding to the (n−1)^thlight path and port n corresponding to the n^thlight path, wherein each variable is 1 when a packet has arrived and 0 when no packet has arrived and;

input variables c_n−1and c_nthat indicate the priority-class of optical packets arriving at port (n−1) and port n, are each 1 when a packet belongs to high-priority class and 0 when a packet belongs low-priority-class, and perform the following processing:

l₀=0, c₀=0;

for each processor n, in parallel (n=1 to N)

f_n,1=l_n−1+l_n;

g_n,1=c_n−1l_n−1+c_nl_n;

and provide output of f_n,1as the number of packets arriving at the two ports and output of g_n,1as a number of high-priority-class packets arriving at the two ports.

20. The buffering device for optical packets according to claim 19, wherein

each of the respective processors p(1,n) of the arrangement of the first stage is provided with a register R(1,n) that temporarily stores output (f_n,1, g_n,1) of the respective processors p(1,n), and

each of the respective processors of the arrangement of the (m−1)^thstage are provided for the integers m from 2 to log₂N;

the processor p(m, n) in the arrangement of the m^th-stage, for the integers n from β+1 to N where β=2^m−1uses the value of the n^thregister R(m−1,n)connected to the processor p(m−1,n) in the arrangement of the respective previous-stage and the value of the (n−β)^thregister R(m−1,n−β) connected to processor p(m−1,n−β) and performs the processing

f_n,m=f_n−m−1+f_n−β,m−1

g_n,m=g_n−m−+g_n−β,m−1

and provides outputs of f_n,mand g_n,m.

21. The buffering device for optical packets according to claim 20, wherein, M=log₂N and

the processors p(M+1, n) (where 1≦n≦N) in the arrangement of the (m+1)^th-stage use

f_n−1,Mas the number of optical packets arriving from ports 1 to n−1,

g_n−1,Mas the number of high-priority-class optical packets arriving from ports 1 to n−1,

l_nwhich indicates the presence of an optical packet arriving at port n,

c_nwhich indicates the priority-class information of an optical packet arriving at port n, and

when l_n=1, perform the following processing

1) when q is less than a predetermined threshold TH, set Δ_n=q+f_n−1,M;

when q is not less than TH, set Δ_n=q+g_n−1,M;

2) when Δ_nis less than B and when c_nshows a high-priority, or, when q is less than TH and when Δ_nis less than B and when c_nshows a low-priority, the optical packet arriving at port n is given delay Δ_n;

and when Δ_nis not less than B or when c_nshows a low-priority and when Δ_nis not less than TH, the optical packet arriving at port n is discarded;

and outputs the value of the delay Δ_nas information that specifies the delay element.

22. The buffering device for optical packets according to claim 21, wherein, for the f_N,Mand g_N,Mstored in register R(M,N) and for q stored in register q,

when q is less than said threshold TH, the larger value between (min(q+f_n,M,B)−1) and 0 is stored as the new queue length (q_new) in register q,

and when q is not less than said threshold TH, the larger value between (min(q+g_n,M,B)−1) and 0 is stored as the new queue length (q_new) in register q.

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to an optical packet buffering device that uses optical circuits that have limited buffer capacity or data holding time, in order to guide a plurality of optical packets such that they are not overlapped toward one lightpath, and to the buffering method thereof.

2. Description of the Prior Art

With advances in optical communications technology able to transfer large amounts of information, in order to achieve even higher levels of throughput, optical technology is being introduced into not only the transmission paths but also into optical packet switching. Typically in optical packet switching, an optical packet transfer process is performed and an optical packet switch is used therein. The functions of such an optical packet switch can be divided into the five categories of: (1) label lookup, (2) switching, (3) buffer management, (4) buffering and (5) routing. Typically, an optical packet switch that implements functions (2) and (4) above optically is called an optical packet switch. This consists of optical switches and optical buffers, having broadband characteristics because the data is transferred while remaining as an optical signal.

In addition, by improving (3) buffer management above, it is possible to make further improvements to the throughput of an optical packet switch. Logic circuits based on electronic circuits had conventionally been used for this (3) buffer management. This is because practical optical logic circuits had not been available.

One of the objects of buffer management is to prevent the collision of optical packets. In an optical packet switching network, if a plurality of optical packets should arrive simultaneously at an optical packet switch, then the optical packets are stored in different positions of buffers to avoid collision. FIG. 1 shows a queue that represents the state of a buffer. For example, in the current Internet technology, packet transmission is done without performing priority control as shown in FIG. 1. The packets that arrive at the packet switch are stored in the buffer if space is available regardless of priority.

In addition, an optical packet switch does not use a semiconductor memory, which is often used in the buffer described above. Since an optical packet switch does not use semiconductor memory for buffering, a method of implementing optical buffer must be used as reported in the literature such as Document 1: K. Habara et al., “Development of a large-capacity photonic packet switch using wavelength routing techniques,” Technical Report of IEICE, SSE99-149, February 2000, and Document 2: David K. Hunter and Meow C. Chia and Ivan Andonovic, “Buffering in Optical Packet Switches,” IEEE/OSA Journal of Lightwave Technology, Vol. 16, No. 12, pp. 2081-2094, December 1998.

Previously known optical buffers have used a plurality of optical fibers of different lengths. These are optical delay elements with different delay times, so the optical packets are stored for the length of this delay time. A method of controlling the optical packets that are allocated to these optical delay lines to be able to avoid collision is known. At this time, it is not possible to extract an optical packet at an arbitrary time, so it is necessary to prevent the collision of optical packets at the time of output, and for this reason, some sort of buffer management technique must be used at the time of input to the optical buffer, thereby controlling the delay times,

In addition, as examples of the conventional optical packet switch architecture, we shall present N-input×N-output optical packet switches with configurations consisting of N 1-input×N-output buffer-less optical switches and N N-input×1-output buffers. The respective 1-input×N-output switches and N-input×1-output buffers are connected in a substantially lattice-type manner. In order to configure the above 1-input×N-output optical packet switches as N-input×N-output optical packet switches that exhibit adequate performance, it is also necessary to process a plurality of optical packets that arrive simultaneously in a sufficiently short period of time, and thus it is important to adopt a buffer management scheme requiring a small amount of computation.

The above optical packet switches are constituted using optical switches. Known optical switches that can be used in this application include: mechanical optical switches that use drive coils as shown in the schematic diagram in FIG. 3(a), optical switches that use optical semiconductor devices as shown in the block diagram in FIG. 3(b) and Mach-Zehnder type optical switches (not shown).

With the above 1-input×N-output buffer-less optical switches, it is possible to perform high-speed label lookup by means of an optical label lookup function as recited in Document 4: N. Wada, H. Harai, W. Chujo, and F. Kubota, “Photonic packet routing based on ultiwavelength label switching using fiber Bragg gratings,” ECOC 2000 Technical Digest (26^thEuropean Conference on Optical Communication), vol. 4 (No. 10.4.6), pp. 71-72, September 2000. As a result, N-input×N-output optical packet switches exhibit high node throughput. However, in order to avoid collision of optical packets and decrease the percentage discarded, a buffer is required, so an N-input×1-output buffer is connected to each output port of the N-input×N-output optical packet switches.

Here follows a more specific description of the buffer management method for processing variable-length packets. This method is already well known. For example, in Document 3: F. Callegati, “Optical buffers for variable length packets,” IEEE Communications Letters, vol. 4, pp. 292-294, September 2000, control is done by the following method with the addition of round-robin scheduling.

With the output buffer type N-input×N-output optical packet switch shown in FIG. 2, N N-input×1-output buffer and N 1-input×N-output switches are used. When the amount of data transmitted increases, t is possible for up to N optical packets to arrive simultaneously at the N-input×1-output buffer.

In order to prevent the collision (or overlap) of these optical packets, delay line buffers are used. In this case, in order to store all of the optical packets, the process of finding the delay of N optical packets must be performed within the period of time l_minequivalent to the minimum optical packet length. Specifically, the time permitted for processing one optical packet is l_min/N. If this time is made to be equivalent to the processing time for one optical packet when buffering is performed according to a simple round-robin scheme, then this indicates that the buffer management can be performed by means of round-robin scheduling.

FIG. 4(a) shows the structure of a 4-input×1-output optical buffer. In this case, N=4. An optical buffer made up of optical delay elements consists of B(=4) optical delay elements using fiber delay lines, N×(B+1) optical switches and a buffer management device (or buffer manager). The B fiber delay lines {d₀, d₁, d₂, d₃} have lengths of integral multiples of the unit length D of delay line, giving lengths of {0, D, 2D, 3D}. In addition, round-robin scheduling based on a clock signal in the buffer manager is repeated every period of time T. For this reason, the buffer manager receives the optical packet information at a point in time separated from the N×(B+1) optical switch by more than the period of time T, and must use this information to find the delay time required for each of the N optical packets during this predefined period of time T. Since it is necessary to process the sequentially arriving optical packets during the period of time T, the relation T≦l_minmust be satisfied. For example, as shown in FIG. 4(a), four optical packets arrive and the buffer manager stores the optical packets A, B and C in the delay lines d₀, d₂and d₃and discards optical packet D. This type of control is achieved by the use of the buffer manager driving the optical switch. FIG. 4(b) shows the positions of the optical packets, when viewed from the output port, immediately after optical packet A is switched to delay line d₀. As shown in FIG. 4(c), optical packets A, B and C are provided as output without collision. In this example, optical packet B had arrived in the buffer prior to optical packet A, but the arriving optical packets are processed sequentially in a round-robin fashion, so they are output in the order shown in FIG. 4(c).

Here follows a detailed description of the behavior of the buffer manager according to the flowchart in FIG. 5. Table 1 presents the main parts of the program.

[Table 1]

	TABLE 1

	ln	: packet length
	tn (n=1, . . . , N)	: arrival gap time
	f	: temporary variable for queue length
	T	: period
	D	: unit length of the delay element
	B	: number of the delay elements
	for n = 1 to N do
	begin
	if (f! = 0)
	begin
	$\begin{matrix} Δ_{n} = ⌈ \frac{f - t_{n}}{D} ⌉; \\ if Δ_{n} < B \end{matrix}$
	begin
	f = t_n+ l_n+ Δ_nD;
	Packet n is given delay Δ_n;
	end
	else Packet n is discarded;
	end
	end
	f = max(f − T, 0);

The buffer manager performs processing in order to avoid optical packet collisions both internally or upon output. Here, f is defined to be the amount of time from a predetermined start time to within the round-robin repeat time T until the time t_fwhen all of the optical packets within the optical fibers have left the buffer. Hereinafter this f is called the buffer occupancy. Each period T, the buffer manager finds the delay time of the optical packets that have arrived at all ports. The delay times given to the optical packets arriving within the same period are found in the order port k=1, 2, . . . , N. Let us assume that an optical packet with an optical packet length l_karrives at port k at the time t_ktaking the time t₀to be the origin. The, delay time t to be given to the optical packet in order to avoid a collision is t=f−t_k. Here, upon considering that the delay line buffer has a discrete time characteristic, one can see that the optical packets should be given the following time delay (F1 of FIG. 5).

$\begin{matrix} Δ_{k} D = D \times Ceil (\frac{f - t_{k}}{D}) = D \times ⌈ \frac{f - t_{k}}{D} ⌉ & [Equation 1] \end{matrix}$

Here, the function Ceil(x) represents a function that gives the smallest integer that is no less than x. If Δ_k<B, then that packet is stored in the Δ_k^thdelay line (F2). Conversely, if Δ_k≧B, then that optical packet is discarded. In this manner, a portion of the optical packets is discarded and the remainder is stored, so it is necessary to update the buffer occupancy f accordingly. Thus, when an optical packet is stored (F₄), f is updated as follows.
f=t_k+l_k+Δ_kD [Equation 2]

Once the delay of the optical packets at all ports is found, excluding the discarded ports, the buffer occupancy f is updated to the new value of max((f−T),0), thus getting ready for the optical packet processing in the next round-robin period after a time of T.

When such an optical delay line buffer is used, the interval between the optical packet arriving at port k and the optical packet stored immediately previously, or namely the optical packet output at the time indicated by the f used at the time of finding the delay in the optical packet arriving at port k, is as follows.
Θ_k=Δ_kD+t_k−f [Equation 3]

Document 6 (H. Harai, N. Wada, F. Kubota, and W. Chujo, “Contention resolution using multi-stage fiber delay line buffer in a photonic packet switch,” Proceedings of IEEE ICC 2002) recites a method wherein delay line buffer units are disposed in a tree structure, and buffer managers are placed in each. Taking N to be the number of ports, this increases the processing speed by N^1/2times in comparison to the round robin processing described above, but a group of delay lines must be placed at each buffer unit so the increased size of the optical element is a problem. In addition, Document 5 (A. Prakash, S. Sharif, and A. Aziz, “An O(log²N) parallel algorithm for output queuing,” Proceedings of IEEE Infocom 2002.) recites a method of performing the process of storing packets, not in optical buffers but rather in semiconductor memory buffers, as parallel processing. This makes the scale of the managing device N times that when using the round-robin process, and the processing speed is log₂N times as fast.

The round-robin scheme described above proceeds sequentially, with data processing regarding buffering being performed in a predetermined port order, so in order to perform high-speed buffering, it is necessary to perform the data processing at high speed but thus the buffering speed is limited. In addition, when the same number of processor units as the number of lightpaths are provided in parallel as recited in Document 5 above, or when delay line buffer units are provided in a tree structure as recited in Document 6 above, even in a method wherein parallel processing is used, the maximum processing time per processor units increases rapidly as the number of input lightpaths increases.

The present invention has as its object to improve the speed of buffering performed during the parallel processing of data, and provide an optical packet buffering device that is able to avoid a rapid increase in the maximum processing time per arithmetic operator even if the number of inputs increases.

SUMMARY OF THE INVENTION

In order to achieve the aforementioned object, the present invention relates to a buffering device for asynchronously-arriving variable-length optical packets that uses a large number of processor units to improve the throughput of processings, and is characterized in being:

a device that accepts input of asynchronously-arriving variable-length optical packets transmitted over a plurality of lightpaths, and outputs same to a single optical path, comprising:

reading means that reads the optical packet length of variable-length optical packets on the respective lightpaths, reading means that reads a time difference-between start time of a period provided by a clock and arrival time of the subsequent most recent variable-length optical packet (hereafter the time difference is called “arrival gap time”), a queue consisting of a plurality of delay elements that temporarily store the plurality of optical packets and give each optical packet one of B different delay times, an optical switch that directs the optical packet to one of the plurality of delay elements, an optical multiplexer connected by lightpaths to the plurality of delay elements, and an output lightpath connected by a lightpath to the multiplexer, and

a controller that uses the delay times of the delay elements and the optical packet length and the arrival gap time thus read to select the delay element used for temporary storage in order to prevent the optical packets from overlapping in the output lightpath, and

the device being a buffering device for asynchronously-arriving optical packets wherein:

this controller is such that, when the plurality of lightpaths are given sequential numbers n from 1 to N, and the value is set to zero for those in which there is no corresponding lightpath,

for the optical packet length and the arrival gap time in the respective lightpaths,

as the arrangement of the first stage of processors and registers, processors P(1,n) that perform the calculation of the optical packet length and the arrival gap time on the (n−1)^thlightpath and the optical packet length and the arrival gap time on the n^thlight path, and registers R(1,n) that store the results of calculation are disposed,

as the arrangement of the m^th(2≦m≦log₂N) stage, processors P(m,n) that perform the calculation of the registers R(m−1,n−2^m−1) or copies thereof and the registers R(m−1,n) or copies thereof, and registers R(m,n) that store the results of calculation are disposed,

on the (M+1)^thstage where 2^M=N a processor and a register Q that stores the information at the tail end of the queue length are disposed,

moreover, processors P(M+1,1) that determine the delay applied to the first variable-length optical packet from the optical packet length and the arrival gap time for the first lightpath or a copy thereof and the value of the register Q, and

for values of n greater than 1, processors P(M+1,n) that determine the delay applied to the n^thvariable-length optical packet, by calculation on the optical packet presence information for the n^thlightpath or a copy thereof and the values of the register R(M,n−1) and the register Q, are disposed,

and thus the delay time applied to the variable-length optical packets of the respective lightpaths is coordinated based on the output of the processors P(M+1,1) and the processors P(M+1,n).

In addition, this invention pertains to a buffering method for optical packets used in the above buffering device for optical packets comprising steps wherein:

taking N to be the number of the lightpaths and n to be the sequential number,

the reading means takes readings at a period T, and regarding the optical packets in the n^thlight path passing through this reading means at a specified time, the length of the optical packet is l_nand the arrival gap time is t_n, and for n=0, l_n−t_n=0,

when a constitution is used such that the delay of the plurality of the delay elements can be arranged as delays that increase by the delay time D,

each of the processors is provided on one of the respective N lightpaths on the input side, and performs the following processing on the integers n from 1 to N:

1) when l_n−1=0 is true, output t_nas t_n,1, or

- when l_n−1=0 is not true, output t_n−1as t_n,1,
  2) in addition, taking g_n,1=l_n−1+t_n−1,
  3) when l_n=0 is true, output g_n−1as f_n,1, or
- when l_n=0 is not true, taking Δ_nto be the smallest integer no less than (g_n−1−t_n)/D set (t_n+l_n+Δ_nD) to the output f_n,1.

In addition, this invention includes, when the plurality of processors are called first stage processors,

each of the respective first-stage processors P(1,n) is provided with a register R(1,n) that temporarily stores its respective output of (f_n,1, t_n,1), and

for the integers n from 3 to N, calculates the content of the n^thregister and the content of the (n−2)^thregister and gives the output of (f_n,2, t_n,2), where:

1) when f_n−2,1=0 is true, output t_n,1as t_n,2, or
- when f_n−2,1=0 is not true, output t_−2,1as t_n,2,
2) in addition, when f_n,1=0 is true, output f_n−2,1as f_{n,2, or}
- when f_n,1=0 is not true,
2-1) taking Δ_nto be the smallest integer no less than (f_n−2,1−t_n,1)/D set (f_n,1+Δ_nD) to the output f_n,2.

In addition, this invention includes, when the plurality of processors are called first-stage processors,

each of the respective first-stage processors P(1,n) is provided with a register R(1,n) that temporarily stores its respective output of (f_n,1, t_n,1), and when the (m−1)^thstage processors are defined for the integers m from 2 to log₂N,

for the m^th-stage processors, and for the integers n from β+1 to N where β=2^m−1, the m^th-stage (n−β)^thprocessor P(m,n) calculates the content of the n^thregister R(m−1,n) connected to the previous-stage processor P(m−1,n) and the content of the (n−β)^thregister R(m−1,n−β) and gives the output of (f_n,m, t_n,m), where:

1) when f_n−β,m−1=0 is true, output t_n,m−1as t_n,m, or
- when f_n−β,m−1=0 is not true, output t_n−β,m−1as t_n,m,
2) in addition, when f_n,m−1=0 is true, output f_n−β,m−1as f_n,m, or
- when f_n,m−1=0 is not true,
2-1) taking Δ_nto be the smallest integer no less than (f_n−β,m−1−t_n,m−1)/D set (f_n,m−1+Δ_nD) to the output f_n,m.

In addition, the method of determining the value Δ_nfor controlling the delay time according to this invention includes:

for the integers n from 1 to N when the length of the queue is q, and M=log₂N,

1) when l_n≠0 is not true, end processing, and
- when l_n≠0 is true, for f_n−1,Min the register R(M,n−1),
2) when f_n−1,M=0 is true, provide a temporary value (temp), and set (temp)=q,
- when f_n−1,M=0 is not true, set the sum of f_n−1,Mand the smallest integer no less than (q−t_n,−1,M)/D to (temp),
3) taking Δ_nto be the smallest integer no less than ((temp)−t_n)/D,
- when Δ_nis less than B, output Δ_nas information that specifies a delay element,
- but if not less, discard the corresponding optical packet.

In addition, the updating of the queue length according to this invention includes:

1) when f_N,M=0 is true for the f_N,Mstored in register R(M,N), the larger of (q−T) or 0 is stored as the new queue length (q_new) in the register Q,
- when f_N,M=0 is not true, Δ is set as the smallest integer no less than (q−t_N,M)/D and at this time,
2) when this Δ is smaller than the B, store the value of the larger of 0 and (f_N,M+ΔD−T) in the register Q,
- when this Δ is not smaller than B, take (q−T) as the new queue length and store it in the register Q.

In addition, the updating of the queue length according to this invention includes:

if the MTU (maximum transfer unit) is set,

1) when f_N,M=0 is true for the f_N,Mstored in register R(M,N), the larger of (q−T) or 0 is stored as the new queue length (q_new) in the register Q,
- when f_N,M=0 is not true, Δ is set as the smallest integer no less than (q−t_N,M)/D and at this time,
2) when this Δ is smaller than the B, store the value of the larger of 0 and the value found by subtracting T from the lesser of (f_N,M+ΔD) and (BD+MTU) in the register Q as the new queue length,
- when this Δ is not smaller than B, take (q−T) as the new queue length and store it in the register Q.

In addition, this invention pertains to a buffering method for optical packets in the buffering apparatus for optical packets wherein:

taking N to be the number of the lightpaths, n to be the lightpath number from 1 through N, and the period T to be set as a power of 2 times an appropriately set time unit (τ), or namely T−2^aτ (where a is a nonnegative integer), while T is to be no greater than the minimum packet length represented as the length of an integral multiple of the time unit (τ), the reading means performs reading at each period T, and for the optical packets of the n^thlightpath that pass through this reading means at a specified time, the length of the optical packet is l_nrepresented as a length which is an integral multiple of the time unit, the arrival gap time is expressed as t_nwhich is an integral multiple of the length of the time unit, and the optical packet length is expressed as an integral multiple of the unit time that is no less than its length, the arrival gap time is expressed as an integral multiple of the unit time that is no greater than the lead time, and for n=0, l_n=t_n=0,

- for the delay of the plurality of delay elements, in a constitution where the delays can be arranged such that they increase by the delay time D, D is given as 2^k(where k is a nonnegative integer), and, as the method of representation, the integer corresponding to the least significant k bits of the binary representation of the integer x is given as x⁽²⁾, and the binary number consisting of the remaining most significant bits after removing the aforementioned least significant k bits is given as x⁽¹⁾, each of the processors is provided on one of respective N lightpaths on the input side, and performs the following processing on the integers n from 1 to N.
  1) when l_n−1=0 is true, output t_nas t_n,1, or
- when l_n−1=0 is not true, output t_n−1as t_n,1,
  2) in addition, taking g_n,1=l_n−1t_n−1,
  3) when l_n=0 is true, output g_n=1as f_n,1, or
- when l_n=0 is not true, when g_n−1is less than t_n, set g_n−1⁽¹⁾to Δ_n,
- and when g_n−1is not less than t_n, set (g_n'1⁽¹⁾+1) to Δ_n, and
  4) set (t_n⁽²⁾+l_n+Δ_nD) to the output f_n,1.

In addition, the buffering method for optical packets according to this invention includes: each of the respective first-stage processors P(1,n) is provided with a register R(1,n) that temporarily stores its respective output of (f_n,1, t_n,1), and when the (m−1)^thstage processors are defined for the integers m from 2 to log₂N,

for the m^th-stage processors, and for the integers n from β+1 to N where β=2^m−1, the m^thstage (n−β)^thprocessor P(m,n) calculates the content of the n^thregister R(m−1,n) connected to the previous-stage processor P(m−1,n) and the content of the (n−β)^thregister R(m−1,n−β) and gives the output of (f_n,m, t_n,m), where:

1) when f_n−β,m−1=0 is true, output t_n,m−1as t_n,m, or
- when f_n−β,m−1=0 is not true, output t_n−β,m−1as t_n,m,
2) in addition, when f_n,m−1=0 is true, output f_n−β,m−1as f_n,m, or
- when f_n,m−1=0 is not true,
2-1) when f_n−β,m−1⁽²⁾is less than or equal to t_n,m−1⁽²⁾, set f_n−β,m−1⁽¹⁾to Δ_n,
2-2) when f_n−β,m−1⁽²⁾is not less than or equal to t_n,m−1⁽²⁾, set (f_n−β,m−1⁽¹⁾+1) to Δ_n,
3) set (f_n,m−1+2^kΔ_n) to the output f_n,m.

In addition, the buffering method for optical packets according to this invention includes: for the integers n from 1 to N when the length of the queue is q, assuming that M=log₂N,

1) when l_n≠0 is not true, end processing,
2) when l_n≠0 is true,
2-1) when f_n−1,M=0 is true and,
- when q⁽²⁾is less than or equal to t_n⁽²⁾, set q⁽¹⁾to Δ_n,
- when q⁽²⁾is not less than or equal to t_n⁽²⁾, set (q⁽¹⁾+1) to Δ_n,
2-2) when f_n−1,M=0 is not true and,
- when q⁽²⁾is less than or equal to t_n−1,M⁽²⁾, set q⁽¹⁾to Δ₁,
- when q⁽²⁾is not less than or equal to t_n−1,M⁽²⁾, set (q⁽¹⁾+1) to Δ₁, and moreover,
2-3-1) when f_n−1,M⁽²⁾is less than or equal to t_n⁽²⁾, set (Δ₁+f_n−1,M⁽¹⁾) to Δ_n,
2-3-2) when f_n−1,M⁽²⁾is not less than or equal to t_n⁽²⁾, set (Δ₁+1+f_n−1,M⁽¹⁾) to Δ_n,
3) when Δ_nis less than B, output Δ_nas information that specifies a delay element,
- but if not less, discard the corresponding optical packet.

In addition, the buffering method for optical packets according to this invention includes: assuming that M=log₂N,

1) when f_N,M=0 is true,
1-1) when q is greater than T, (q−T) is set as the new queue length (q_new),
1-2) when q is not greater than T, q_new=0 is set,
2) when f_N,M=0 is not true,
2-1) when q⁽²⁾is less than or equal to t_N,M⁽²⁾, q⁽¹⁾is set as Δ,
- when q⁽²⁾is not less than or equal to t_N,M⁽²⁾, (q⁽¹⁾+1) is set as Δ,
3-1) when this Δ is smaller than B and,
3-1-1) when f_N,M+2^kΔ is greater than T, (f_N,M+2^kΔ−T) is set as q_now,
3-1-2) when f_N,M+2^kΔ is not greater than T, q_new=0 is set,
3-2) when this Δ is not smaller than B, (q−T) is set as q_now.

In addition, the buffering method for optical packets according to this invention includes: assuming that M=log₂N,

1) when f_N,M=0 is true,
1-1) when q is greater than T. (q−T) is set as the new queue length (q_new),
1-2) when q is not greater than T, q_new=0 is set,
2) when f_N,M=0 is not true,
2-1) when q⁽²⁾is less than or equal to t_N,M⁽²⁾, q⁽¹⁾is set as Δ,
- when q⁽²⁾is not less than or equal to t_N,M⁽²⁾, (q⁽¹⁾+1) is set as Δ,
3-1) when this Δ is smaller than B and,
3-1-1) when (f_N,M+2^kΔ) is greater than (2^kB+MTU), (2^kB+MTU−T) is set as q_now,
3-1-2) when (f_N,M+2^kΔ) is not greater than (2^kB+MTU−T) and,
3-1-2-1) when f_N,M+2^kΔ is greater than T, (f_N,M+2^kΔ−T) is set as q_new,
3-1-2-2) when f_N,M+2^kΔ is not greater than T, q_new=0 is set,
3-2) when this Δ is not smaller than B, (q−T) is set as q_new.

As described above, when the period T is 2^p(where p is an integer) times or 2^nktimes the delay time D, this invention simplifies the implementation in hardware by eliminating division, and can further increase the speed of data processing.

In addition, this invention includes a device that accepts input of synchronous fixed-length optical packets from a plurality of lightpaths, gives each of the optical packets one of B different delay times and outputs same to a single optical path, comprising:

reading means that reads the presence of optical packets on the respective lightpaths,

a queue consisting of a plurality of delay elements that temporarily store the plurality of optical packets and give each optical packet one of B different delay times, an optical switch that directs the optical packet to one of the plurality of delay elements, an optical multiplexer connected by lightpaths to the plurality of delay elements, and an output lightpath connected by a lightpath to the multiplexer, and

a controller that uses the delay times of the delay elements and the presence information for optical packets thus read to select by computation the delay element used for temporary storage in order to prevent the optical packets from overlapping in the output lightpath, the controller being such that:

when the plurality of lightpaths are given sequential numbers n from 1 to N, and cases in which there is none corresponding to n can be neglected,

for the number of optical packets in the respective lightpaths,

as the arrangement of the first stage of processors and registers, processors P(1,n) that perform the calculation using the optical packet presence information on the (n−1)^thlightpath and the optical packet presence information on the n^thlight path, and registers R(1,n) that store the results of calculation are disposed,

for the integers M where 2^M=N, on the (M+1)^thstage a processor and a register Q that stores the information at the tail end of the queue length are provided,

moreover, processors P(M+1,1) is used to that determine the delay applied to the first synchronous fixed-length optical packet from the optical packet presence information for the first lightpath or a copy thereof and the value of the register Q, and

for values of n greater than 1, processors P(M+1,n) are used to determine the delay applied to the n^thsynchronous fixed-length optical packet, by calculation on the optical packet presence information for the n^thlightpath or a copy thereof and the values of the register R(M,n−1) and the register Q, thereby coordinating the delay times given to the synchronous fixed-length optical packets of the respective lightpaths.

In addition, the buffering device for optical packets according to this invention, further comprising reading means that takes readings of the presence of optical packets in the respective light paths, wherein

the respective first-stage processors receive input variables l_n−1and l_nthat indicate the presence of optical packets arriving at port (n−1) and port n, namely, the variable is 1 when a packet has arrived and 0 when no packet has arrived, and perform the following processing:

l₀:=0

for each processor n, in parallel (n:=1 to N)

f_n,1:=l_n−1+l_n;

and provide output of f_n,1, the number of packets arriving at the two ports.

In addition, this invention includes a device that accepts input of synchronized fixed-length optical packets, gives different delay times to the respective optical packets and outputs same on a single light path, wherein:

each of the respective first-stage processors P(1,n) is provided with a register R(1,n) that temporarily stores its respective output of (f_n,i), and when the (m−1)^thstage processors are defined for the integers m from 2 to log₂N,

for the m^th-stage processors, and

for the integers n from β+1 to N where β=2^m−1, for the value of the n^thregister R(m−1,n) connected to the previous-stage processor P(m−1,n) and the value of the (n−β)^thregister R(m−1,n−β, processor P(m,n) performs the processing:

f_n,m:=f_n−m−1+f_n−β,m−1

and provides output of f_n,m.

the (M+1)^thprocessors P_M+1,n(where 1≦n≦N) use

f_n−1,Mas the number of optical packets arriving from ports 1 to n−1,

l_nwhich indicates the presence of an optical packet arriving at port n, and

the queue length q, for the integers n from 1 to N, and perform the processing:

if l_n=0 exit;

else Δ_n:=q+f_n−1,M;

but if Δ_n<B Packet n is given delay Δ_n;

and if not Δ_n<B Packet n is discarded;

in this order, and outputs the value of the delay Δ_nas information that specifies the delay element.

for the f_N,Mstored in register R(M,N), the value of the larger of (min(q+f_n,M,β)−1) and 0 is stored as the new queue length (q_new) in register Q.

As described above, since arrival gap time is always zero and packet length is unity in the case of synchronously-arriving fixed-length packets, this invention simplifies the calculation and the implementation in hardware, and can further increase the speed of data processing.

reading means that reads the presence of optical packets and priority-class information of the optical packets on the respective lightpaths,

and a controller that selects the delay element to be used for temporary storage for the optical packets in order to prevent the optical packets from overlapping in the output lightpath by using the delay times of the delay elements and the presence information of the optical packets and the priority-class information of the optical packets,

and this controller is such that:

when the plurality of lightpaths are given sequential numbers n from 1 to N, and in the following, cases in which there is none corresponding to n can be neglected,

for the presence information of the optical packets and the priority-class information of the optical packets in the respective lightpaths,

as the arrangement of the first stage of processors and registers, processors P(1,n) that perform the calculation using the optical packet presence information and priority-class information on the (n−1)^thlightpath and the optical packet presence information and priority-class information on the n^thlight path, and registers R(1,n) that store the results of calculation are disposed,

for the integers M where 2^M=N, on the (M+1)^thstage a processor and a register Q that stores the information at the queue length are provided,

moreover, processor P(M+1,1) is used to determine the delay applied to the first synchronous fixed-length optical packet from the optical packet presence information and the optical packet priority-class information for the first lightpath or a copy thereof and the value of the register Q, and

for values of n greater than 1, processors P(M+1,n) are used to determine the delay applied to the n^thsynchronous fixed-length optical packet, by calculation on the optical packet presence information and optical packet priority-class information for the nth lightpath or a copy thereof and the values of the register R(M,n−1) and the register Q, thereby coordinating the delay times given to the synchronous fixed-length optical packets of the respective lightpaths.

In addition, this invention includes a device that further comprises reading means that takes readings of the presence of optical packets and the priority-class information of optical packets in the respective light paths, wherein

the respective first-stage processors receive input variables l_n−1and l_nthat indicate the presence of optical packets arriving at port (n−1) and port n of 1 when a packet has arrived and 0 when no packet has arrived, receive input variables c_n−1and c_nthat indicate the priority-class of optical packets arriving at port (n−1) and port n of 1 when a packet belongs to high-priority class and 0 when a packet belongs low-priority-class, and perform the following processing:

l₀:=0, c₀:=0;

for each processor n, in parallel (n:=1 to N)

begin

f_n,1:=l_n−1+l_n;

g_n,1:=c_n−1l_n−1c_nl_n;

end,

and provide output of f_n,1, the number of packets arriving at the two ports and output of g_n,1, the number of high-priority-class packets arriving at the two ports.

In addition, this invention includes a device that each of the respective first-stage processors P(1,n) is provided with a register R(1,n) that temporarily stores its respective outputs of (f_n,1, g_n,1), and when the (m−1)^thstage processors are defined for the integers m from 2 to log₂N,

for the m^th-stage processors, and

f_n,m:=f_n−m−1+f_n−β,m−1

g_n,m:=g_n−m−1+g_n−β,m−1

and provides outputs of f_n,mand g_n,m.

In addition, this invention includes a device that the (M+1)^thprocessors P_M+1,n(where 1≦n≦N) use

f_n−1,Mas the number of optical packets arriving from ports 1 to n−1,

g_n−1,Mas the number of high-priority-class optical packets arriving from ports 1 to n−1,

l_nwhich indicates the presence of an optical packet arriving at port n,

c_nwhich indicates the priority-class information of an optical packet arriving at port n, and

the queue length q, for the integers n from 1 to N, and when l_n=1, perform the following processing:

1) when q is less than the predetermined threshold TH, set Δ_n:=q+f_n−1,M;

- when q is not less than TH, set Δ_n:=q+g_n−1,M;
  2) when Δ_nis less than B and when c_nshows high-priority, or, when q is less than TH and when Δ_nis less than B and when c_nshows low-priority class, Packet n is given delay Δ_n;
  and when Δ_nis not less than B or when c_nshows low-priority and when Δ_nis not less than TH, Packet n is discarded;
  in this order, and outputs the value of the delay Δ_nas information that specifies the delay element.

In addition, this invention includes a device for the f_N,Mand g_N,Mstored in register R(M,N) and for q stored in register Q,

when q is less than the threshold TH, the larger value between (min(q+f_n,M,B)−1) and 0 is stored as the new queue length (q_new) in register Q,

and when q is not less than the threshold TH, the larger value between (min(q+g_n,M,B)−1) and 0 is stored as the new queue length (q_new) in register Q.

BRIEF EXPLANATION OF THE DRAWINGS

FIG. 1 is a diagram illustrating a queue that represents the state of a buffer that is typically used.

FIG. 2 is a diagram illustrating an example of conventional optical packet switch architecture.

FIG. 3 illustrates an optical switch that can be used in an optical packet switch, where FIG. 3(a) shows a mechanical type of optical switch, and FIG. 3(b) shows an optical switch that uses semiconductor devices.

FIG. 4 is a diagram illustrating the construction of a 4-input×1-output optical buffer, where FIG. 4(a) shows the state when four optical packets arrive, FIG. 4(b) shows the position of optical packet A when switched to delay line d₀, and FIG. 4(c) shows the state when the three optical packets A, B and C are output.

FIG. 5 is a flowchart of the processing of the round-robin scheduling.

FIG. 6 is a diagram illustrating the optical packet switch including buffering device according to the present invention.

FIG. 7 is a block diagram illustrating the constitution of an N×1 buffer,

FIG. 8 is a block diagram illustrating the optical packet switch including optical buffering device according to the present device.

FIG. 9 is a block diagram illustrating the constitution of an N×1 buffer.

FIG. 10(a) is a block diagram illustrating an example of the management information processor of the device of FIG. 8, and FIG. 10(b) is a block diagram illustrating another example of a management information processor.

FIG. 11(a) is a block diagram illustrating the constitution of the management information processor of the device of FIG. 8, and FIG. 11(b) is a block diagram illustrating another example of a management information processor.

FIG. 12 is a time chart for describing a buffering device for asynchronous variable-length optical packets, where FIG. 12(a) shows the state when four optical packets arrive, FIG. 12(b) shows the position of optical packet A when switched to delay line d₀, and FIG. 12(c) shows the state when the three optical packets A, B and C are output.

FIG. 13 is a block diagram of a parallel processing circuit used in the case of asynchronous variable-length optical packets.

FIG. 14 is a time chart for describing a buffering device for asynchronous variable-length optical packets.

FIG. 15 is a flowchart of the processing of the first parallel processor.

FIG. 16 is a flowchart of the processing of the second parallel processor.

FIG. 17 is a flowchart of the processing of the third parallel processor.

FIG. 18 is a flowchart of the processing of the m^thparallel processor.

FIG. 19 is a flowchart of the processing of the parallel processor at the stage of output of the delay time.

FIG. 20 is a flowchart of the processing of the parallel processor performing an update of the queue length.

FIG. 21 is a flowchart of the processing of the first parallel processor.

FIG. 22 is a flowchart of the processing of the second parallel processor.

FIG. 23 is a flowchart of the processing of the third parallel processor.

FIG. 24 is a flowchart of the processing of the parallel processor at the stage of output of the delay time.

FIG. 25 is a flowchart of the processing of the parallel processor performing an update of the queue length.

FIG. 26 is a block diagram of another parallel processing circuit used for asynchronous variable-length optical packets.

FIG. 27 is a block diagram of a parallel processing circuit used for synchronous fixed-length optical packets.

FIG. 28 is a block diagram of another parallel processing circuit used for synchronous fixed-length optical packets.

FIG. 29 is a flowchart for finding delay in the case of synchronous fixed-length optical packets.

FIG. 30 is an example of a plot illustrating the relationship of the number of logic elements (y-axis) required to achieve an image circuit that has a specified number of input ports (x-axis).

DESCRIPTION OF THE PREFERRED EMBODIMENT

Here follows a detailed description of a preferred embodiment of the preferred embodiment made with reference to the drawings. In the examples below, devices or the like that have the same or similar functions are given the same numbers.

FIG. 6 illustrates an overview of a buffering device for asynchronously arriving variable-length optical packets. FIG. 6 is a device that accepts input of asynchronously-arriving variable-length optical packets transmitted on a plurality of lightpaths and outputs the variable-length optical packets to an optical path depending on address information in the packet, so optical packet signals transmitted over lightpaths 1 through 4 are directed into outputs 1 through 4. The signals input from each of the respective lightpaths are converted with an optical detector 1 to an electrical signal, and their header information is read and processed by a route information processor 5 as route information. This route information is sent to a (1×N) optical switch 3 where M=4 and thus the output direction is determined. The delay lines 2 between the optical detector 1 and (1×N) optical switches 3 are used to synchronize the optical packet signals and electrical signals.

Here, the (N×1) buffers 4 are buffers characteristic of the present invention, having the constitution described in FIG. 7. Each buffer comprises: reading means that reads the optical packet length of the variable-length optical packets on the respective lightpaths, reading means that reads a time difference (the lead time of the optical packet) between a clock prepared in advance and the subsequent most recent variable-length optical packet, a queue consisting of delay elements that temporarily store the optical packet and give the optical packet one of B different delay times, an optical switch that directs the optical packet to one of the plurality of delay elements, an optical multiplexer connected by lightpaths to the plurality of delay elements, and an output lightpath connected by a lightpath to the multiplexer.

FIG. 7 illustrates the constitution of the (N×1) buffer 4. An optical detector 6 converts the optical signal to an electrical signal so that the buffer manager 11 can thus read the length l of the optical packet signal and the interval (the lead time of the packet) t from the optical packet to a clock signal generated internally or supplied from outside. In this buffer manager 11, these values of l and t are used to perform data processing so that the optical packet signals transmitted from the lightpaths through the respective delay lines 7 do not overlap. This output is used to control an optical switch 8 and determine which of the delay lines 9 is to be selected. The optical packet signals that pass through each of the delay lines 9 are multiplexed by the multiplexer 10 and output.

In the constitution illustrated in FIG. 6, each of the optical packet signals is converted to an electrical signal twice, but as in the constitution of a buffering device for asynchronously-arriving variable-length optical packets shown in FIG. 8, it may be converted to an electrical signal in the optical detector 1 and this electrical signal can be used by the management information processor 12 as is, or after being subjected to data processing, to control the (1×N) optical switch 3 and (N×1) buffers 13. In contrast to the constitution of the (N×1) buffers 4, the constitution of the (N×1) buffers 13 in this case is such that, as illustrated in FIG. 9, it is possible to reduce the number of times the optical-to-electrical conversion is performed and thus reduce the number of optical detectors.

As the management information processor 12, it is possible to use one of the constitution illustrated in FIG. 10(a) or FIG. 11(a).

In the one of FIG. 10(a), the high/low (H/L) voltage of the signal converted to an electrical signal is determined and then this is recorded in a time recorder as the time to perform high/low switching; this time data is recorded in a register and this data is periodically supplied to the buffer manager.

In addition, in the one of FIG. 11(a), the high/low voltage of the signal converted to an electrical signal is determined and then this is regularly written to a shift register. The time to perform high/low switching is read as the address of the shift register and compared against the clock signal to obtain the length l of the optical packet signal and the interval t between the optical packets.

A unit of the constitution shown in FIG. 10(b) can be used as the management information processor 12. This unit branches the output of the optical detector, directing one branch to a circuit that generates a signal supplied to the buffer manager and directing the other branch to a circuit that extracts route information. In order to generate a signal to be supplied from one branch to the buffer manager, the high/low voltage of the signal converted to an electrical signal is determined and the time to perform high/low switching is recorded in a time recorder and this time data is recorded in a register. This register outputs information for a certain optical packet signal at a specified time. A switch is used to direct this output to the respective buffer managers according to the output from the route information reader.

A unit of the constitution shown in FIG. 11(b) can be used as the management information processor 12. This unit determines the high/low voltage of the signal converted from an optical packet signal to an electrical signal then this is regularly written to a shift register. The time to perform high/low switching is read as the address of the shift register to obtain the length l of the optical packet signal and the interval t between the optical packets.

Here follows a description of the length l of the optical packet signal at a specified buffering time and the interval t between the optical packet and clock signal. First, the buffering is assumed to be performed every T seconds and this time T is to be smaller than the minimum value of l. Specifically, T is determined so that in the interval from time T₀to time T₁which is T seconds later, no more than one packet arrives at the same port. At this time, if no optical packet signal should arrive in the interval from time t₀to time T₁, then l is zero and t is also zero. However, if an optical packet signal arrives in the interval from time T₀to time T₁, then the length of this optical packet signal is set as l and the interval between the clock signal and the end of the optical packet signal closest to time T₁is set as t. Specifically, in the event that an optical packet signal is so long as to extend across a plurality of periods, then that optical packet signal is deemed to belong to the first period time that it extends across.

The following description pertains to a buffering device as shown in FIG. 12 that coordinates variable-length packets transmitted asynchronously over a plurality of lightpaths so that they do not overlap and are sent together over a single lightpath. In order to prevent overlapping, the length of the arriving optical packet signals and the interval between the optical packet signal and the clock signal are used to determine the delay element to which each of the optical packet signals is to be assigned, thereby giving the optimal delay. In the device of the constitution illustrated in FIG. 12, the constitution is such that the optical packet signals of lightpath 1 are given priority such that their delay is minimized, but it is also possible to give priority to the other light paths, so if there is a lightpath that is to be given priority, then it can be used to replace lightpath 1 of FIG. 12. Here, regarding the constitution of the delay elements, in FIG. 12 these are given equal differences of 0, D, 2D and 3D but even in the case that equal differences are not used, the method and constitution described below can be applied with only minor variations.

FIG. 12(b) is one example of the delays given to lightpaths 1 through 4. The optical packet signal of lightpath 4 (signal D) is discarded, so it does not appear in FIG. 12(b). The optical packet signals that are given delay times and coordinated so that they do not overlap in this manner are lined up as shown in FIG. 12(c). Here, if it is possible to select even longer delay times, then one can easily imagine how the optical packet signal of lightpath 4 (signal D) can be transmitted and not discarded.

Example 1

The main point of the present invention is to determine the delay times for as many packets as possible per time unit. To this end, a large number of processors P_ijare disposed as shown in FIG. 13 to perform the processing required to determine the delay times in parallel. Here, P_ijrefers to the processor provided on the i^thstage of the j^thlightpath. In order to be particularly clear, P_ijmay be represented as P(i,j).

In FIG. 13, the length l (L) of the optical packet signals of the eight lightpaths from IN₁to IN₈and the interval (packet pre-period) t from the optical packet signal to the clock signal are used to perform coordination so that the percentage of optical packets discarded is minimized. In this case also, delay elements that are arranged with equal differences between their delay times are shown, and the Δ_nserving as the output is shown as a coefficient of the base delay time of D. In FIG. 13, the circles indicate processors and the squares are registers R. Here, the registers include registers R(1,1) that store the results of processing by the immediately previous processor and the registers R′(1,1) that store the optical packet length of the optical packet on the first light path and the lead time of the optical packet. The respective registers R_ijalso have their positions indicated by the subscripts (ij) and are also indicated R(i,j), When there is a processor, the same subscripts are used. This register R′(1,1) is used to coordinate the timing of the arithmetic processing, so it is not absolutely necessary to be provided in this portion, but rather processing may be performed with the constitution illustrated in FIG. 26, so it may be sufficient to perform the coordination of this timing using a well-known delay circuit (although not shown in the figure) until the process of updating the position in the queue is performed. However, the circuit design can be simplified by using parallel registers R(1,1) as shown above.

In the constitution shown in FIG. 13, in order to prevent the input optical packets from overlapping in the lightpath used for output, the delay times of the delay elements and the read optical packet lengths and optical packet pre-periods are used to determine the delay element where it is to be temporarily stored. For convenience, we shall distinguish the plurality of lightpaths with a sequence number n from 1 to N. In the following, if there is no processor or register corresponding to the number of a lightpath, then its function is to perform an identity operation or the value it holds is zero,

In this case, the optical packet length and the lead time of the optical packet of each respective lightpath is calculated using a buffering device for optical packets of the following constitution. This is a buffering device for asynchronously arriving optical packets characterized in that it:

1) comprises reading means that reads the optical packet length of the variable-length optical packets on the respective lightpaths, reading means that reads a time difference (the lead time of optical packet) between a clock prepared in advance and the subsequent most recent variable-length optical packet, a queue consisting of delay elements that temporarily store the optical packet and give the optical packet one of B different delay times, an optical switch that directs the optical packet to one of the plurality of delay elements, an optical demultiplexer connected by lightpaths to the plurality of delay elements, and an output lightpath connected by a lightpath to the demultiplexer,

2) comprises a controller that, in order to prevent the input optical packets from overlapping in the lightpath used for output, uses the delay times of the delay elements and the read optical packet lengths and optical packet pre-periods to calculate and determine the delay element where it is to be temporarily stored,

3) this controller is such that, when the plurality of lightpaths are given sequential numbers n from 1 to N, and in the following, the value is set to zero for those in which there is no corresponding lightpath,

for the optical packet length and the optical packet pre-period in the respective lightpaths,

as the arrangement of the first stage of processors and registers, processors P(1,n) that perform the calculation of the optical packet length and the optical packet pre-period on the (n−1)^thlightpath and the optical packet length and the optical packet pre-period on the n^thlight path, and registers R(1,n) that store the results of calculation are disposed,

- 4) as the arrangement of the m^th(2≦M≦log₂A stage, processors P(m,n) that perform the calculation of the registers R(m−1,n−2^m−1) or copies thereof and the registers R(m−1,n) or copies thereof, and registers R(1,n) that store the results of calculation are provided,

5) on the (M+1)^thstage where 2^M=N are provided a processor and a register Q that stores the information at the tail end of the queue length,

6) moreover, processors P(M+1,1) that determine the delay applied to the first variable-length optical packet from the optical packet length and the lead time of the optical packet for the first lightpath or a copy thereof and the value of the register Q, and

7) for values of n greater than 1, processors P(M+1,n) that determine the delay applied to the n^thsynchronous fixed-length optical packet, by calculation on the optical packet presence information for the n^thlightpath or a copy thereof and the values of the register R(M,n−1) and the register Q, thereby coordinating the delay times given to the synchronous fixed length optical packets of the respective lightpaths are provided,

8) and thus the delay time applied to the variable-length optical packets of the respective lightpaths is coordinated based on the output of the processors P(M+1,1) or the processors P(M+1,n).

The functions of the processors P_ijwill be described later. In the case that there are N elements a₁, a₂, . . . , a_N, the prefix-sum operation is known as the way of finding the respective prefix sums a₁, a₁+a₂, . . . , a₁+a₂+, . . . , a_N, and the parallel prefix operation is known as a method of finding the solution by parallel processing using N processors. However, when this method is used to find the delay of packets, it is necessary to perform the calculation log₂N+1 times per processor and the delay of the next packet cannot be found during that period. In addition, the pipeline is known as a constitution that improves throughput by sending the results of each operation sequentially to the next stage, but even if previously known processing methods are applied, they cannot be used in the present invention. Processors of a unique constitution are required for application to the present invention, and processing suited to the buffering device and buffering method is required.

FIG. 14 presents the correspondence among the variables used below on a time chart. FIG. 14(a) is an example of the order of optical packet signals envisioned as the output of the buffering device according to the present invention. The input in this case is given in eight lightpaths. In addition, FIG. 14(b) presents the meaning of the variables used in the following description. In addition, FIG. 14(c) presents the coefficient of the base delay time, and this becomes the output.

The first-stage processors P_1nperform processing according to the flowchart shown in FIG. 15. If this processing were written in the form of a program, it would be as follows.

[Table 2]

	TABLE 2

	First Stage (P_1n: 1≦n≦N)
	l_n, l_n−1	: packet length
	t_n, t_n−1	: arrival gap time
	g_n−1	; temporary variable for queue length
	f_n,1, t_n,1	: output
	T	: period
	D	: unit length of the delay element
	B	: number of the delay elements
	for each processor n, in parallel (n = 1 to N)
	begin
	if (l_n−1= 0) t_n,1= t_n;
	else t_n,1= t_n−1;
	g_n−1= t_n−1+ l_n−1; :
	if (l_n= 0) f_n,1= g_n−1
	else begin
	$\begin{matrix} Δ_{n} = ⌈ \frac{g_{n - 1} - t_{n}}{D} ⌉; \\ f_{n, 1} = t_{n} + l_{n} + Δ_{n} D; \end{matrix}$
	end
	end

In Table 2 above, the length of the optical packet is l_nand the optical packet pre-period is t_n, for n=0, l_n=t_n=0, and the delay of the plurality of delay elements can be arranged as delays that increase by the delay time D. These first-stage processors are provided with N lightpaths on the input side, and perform the following processing on the integers n from 1 to N.

1) when l_n=1=0 is true, output t_nas t_n,1, or

- when l_n−1=0 is not true, output t_n−1as t_n,1,
  2) in addition, taking g_n,1=l_n−1+t_n−1,
  3) when l_n=0 is true, output g_n−1as f_n,1, or
- when l_n=0 is not true, taking Δ_nto be the smallest integer no less than (g_n−1−t_n)/D set (t_n+l_n+Δ_nD) to the output f_n,1.

The second-stage processors P_2nperform processing according to the flowchart shown in FIG. 16. If this processing were written in the form of a program, it would be as follows.

[Table 3]

	TABLE 3

	Second Stage (P_2n: 3≦n≦N)
	fn,1 , fn−2,1	: last position of the combined packet
	t_n,1, t_n-2,1	: arrival gap time
	g_n-1	: temporary variable for queue length
	f_n,2, t_n,2	: output
	for each processor n, in parallel (n = 3 to N)
	begin
	if (f_n−2,1= 0) t_n,2= t_n,1;
	else t_n,2= t_n−2,1;
	if (f_n,1= 0) f_n,2= f_n−2,1;
	else begin
	$\begin{matrix} Δ_{n} = ⌈ \frac{f_{n - 2, 1} - t_{n, 1}}{D} ⌉; \\ f_{n, 2} = f_{n, 1} + Δ_{n} D; \end{matrix}$
	end ;
	end

Table 3 above presents the buffering method of a device provided with the second-stage (n−2)^thprocessor that, for the integers n from 3 to N, calculates the content of the nth register and the content of the (n−2)^thregister and gives the output of (f_n,2, t_n,2). Here, the following processing is performed.

On the integers n from 1 to N.

1) when f_n−2,1=0 is true, output t_n,1as t_n,2, or
- when f_n−2,1=0 is not true, output t_n−2,1as t_n,2,
2) in addition, when f_n,1=0 is true, output f_n−2,1as f_n,2, or
- when f_n,1=0 is not true,
2-1) taking Δ_nto be the smallest integer no less than (f_n−2,1−t_n,1)/D set (f_n,1+Δ_nD) to the output f_n,2.

The third-stage processors P_3nperform processing according to the flowchart shown in FIG. 17. If this processing were written in the form of a program, it would be as follows.

[Table 4]

	TABLE 4

	Third Stage (P_3n: 5≦n≦N)
	fn,2 , fn−1,2	: last position of the combined packet
	t_n,2, t_n−1,2	: arrival gap time
	f_n,3, t_n,3	: output
	for each processor n, in parallel (n = 5 to N)
	begin
	if (f_n−4,2= 0) t_n,3= t_n,2;
	else t_n,3= t_n−4,2;
	if (f_n,2= 0) f_n,3= f_n−4,2;
	else begin
	$\begin{matrix} Δ_{n} = ⌈ \frac{f_{n - 4, 2} - t_{n, 2}}{D} ⌉; \\ f_{n, 3} = f_{n, 2} + Δ_{n} D; \end{matrix}$
	end ;
	end

In general, the m^th-stage processors P_mnperform processing according to the flowchart shown in FIG. 18. If this processing were written in the form of a program, it would be as follows.

[Table 5]

	TABLE 5

	m-th Stage (P_mn: 2^m−1+1≦n≦2^M=) : M=log₂N
	f_n,m−1, f_n−2_m−1_,m−1	: last position of the combined packet
	t_n,m−1, t_n−2_m−1_,m−1	: arrival gap time
	f_n,m, t_n,m	: output
	for each processor n, in parallel (n = 2^m−1+ 1 to N)
	begin
	if (f_n−2_m−1_,m−1= 0) t_n,m= t_n,m−1;
	else t_n,m= t_n−2_m−1_,m−1;
	if (f_n,m−1= 0) f_n,m= f_n−2_m−1_,m−1;
	else begin
	$\begin{matrix} Δ_{n} = ⌈ \frac{f_{n - 2^{k - 1}, k - 1} - t_{n, k - 1}}{D} ⌉; \\ f_{n, k} = f_{n, k - 1} + Δ_{n} D; \end{matrix}$
	end ;
	end

Table 5 above presents the buffering method of a device that, when an (m−1)^thprocessor is defined, is provided with the m^th-stage (n−β)^thprocessor that, for the integers n from β+1 to N where β=2^m−1, calculates the content of the n^thregister and the content of the (n−β)^thregister and gives the output of (f_n,m, t_n,m). Here, the following processing is performed. Assuming that 2≦m≦log₂N.

1) when f_n−β,m−1=0 is true, output t_n,m−1as t_n,m, or
- when f_n−β,m−1=0 is not true, output t_n−β,m−1as t_n,m,
2) in addition, when f_n,m−1=0 is true, output f_n−β,m−1as f_n,m, or when f_n,m−1=0 is not true,
2-1) taking Δ_nto be the smallest integer no less than (f_n−β,m−1−t_n,m−1)/D set (f_n,m−1+Δ_nD) to the output f_n,m.

As data for % electing delay elements that are arranged with equal differences between their delay times of 0, D, 2D, . . . , (B−1)D, when the delay time of the n^thlightpath is given as Δ_nD, output this Δ_n. To this end, the processing according to the flowchart shown in FIG. 19 is performed. If this processing were written in the form of a program, it would be as shown in Table 6. However, this is used in a device wherein each of the first-stage processors P(1,n) is provided with a register R(1,n) that temporarily stores its respective output (f_n,1, t_n,1), and that when an (m−1)^thprocessor is defined for the integers m from 2 to log₂N, is provided with the m^th-stage (n−β)^thprocessor P(m,n) that, for the integers n from β+1 to N where β=2^m−1for the m^thprocessor, calculates the content of the n^thregister R(m⁻¹,n) and the content of the (n−β)^thregister R(m−1,n−β) and gives the output of (f_n,m, t_n,m).

[Table 6]

	TABLE 6

	(m+1)_m−1-th Stage ,Delay time, (P_M+1,n: 1≦n≦N) : M=log₂N
	f_n−1,M_m−1	: last position of the combined packet
	l_n	: packet length arrived M periods before
	t_n	: arrival gap time M periods before
	t_n−1,M	: arrival gap time
	q	: queue length
	Δ_n	: delay
	for each processor n, in parallel (n = 1 to N)
	begin
	if (l_n= 0) exit;
	if (f_n−1,M= 0) temp = q;
	$\begin{matrix} else temp = ⌈ \frac{q - t_{n - 1, M}}{D} ⌉ + f_{n - 1, M}; \\ Δ_{n} = ⌈ \frac{temp - t_{n}}{D} ⌉; \end{matrix}$
	If (Δ_n< B) Packet n is given delay Δ_n;
	else Packet n is discarded;
	end

Table 6 above presents the method of finding Δ_nfor the integers n from 1 to N when the length of the queue is q, by performing the following processing. However, in the following, M=log₂N.

1) When l_n≠0 is not true, end processing, and
- when l_n≠0 is true, for f_n−1,Min the register R(M,n−1),
2) when f_n−1,M=0 is true, provide a temporary value (temp), and set (temp)=q,
- when f_n−1,M=0 is not true, set the sum of f_n−1,Mand the smallest integer no less than (q−t_{n,−1, M})/D to (temp),
3) taking Δ_nto be the smallest integer no less than ((temp)−t_n)/D,
- output Δ_nwhen Δ_nis less than B,
- but if not less, discard the corresponding optical packet.

In this manner, taking M=log₂N, at the same time that the M^th-stage processing results are obtained, the queue length (q) is updated by a separate processor. This processing is performed according to the flowchart shown in FIG. 20. If this processing were written in the form of a program, it would be as follows.

[Table 7]

TABLE 7

(m+1)_m−1-th Stage, Queue length, : M=log2_N
f_N,M	: last position of the combined packet
t_N,M_m−1	: arrival gap time
q	: queue length
Δ	: temporary variable
q_new	: output queue length
	if (f_N,M= 0) q_new= max(q − T, 0);
	else begin
	$\begin{matrix} Δ = ⌈ \frac{q - t_{N, M}}{D} ⌉; \\ if (Δ < B) \end{matrix}$
	q_new= max (min(f_N,M+ ΔD, BD + MTU) − T, 0);
	else q_new= q − T;
	end

Generally, in packet communications, the MTU (maximum transfer unit) is set. Table 7 above presents the procedure in the case that the MTU is set.

1) When f_N,M=0 is true for the f_N,Mstored in register R(M,N), the larger of (q−T) or 0 is stored as the new queue length (q_now) in the register Q,
- when f_N,M=0 is not true, Δ is set as the smallest integer no less than (q−t_N,M)/D and at this time,
2) when this Δ is smaller than B, store the value of the larger of 0 and the value found by subtracting T from the lesser of (f_N,M+ΔD) and (BD+MTU) in the register Q,
- when this Δ is not smaller than B, take (q−T) as the new queue length and store it in the register Q.

If the MTU (maximum transfer unit) is not set, or if the speed of processing is to be increased even if the MTU is present, the queue length can be set by the following method.

1) When f_N,M=0 is true for the f_N,Mstored in register R(M,N), the larger of (q−T) or 0 is stored as the new queue length (q_now) in the register Q,
- when f_N,M=0 is not true, Δ is set as the smallest integer no less than (q−t_N,M)/D and at this time,
2) when this Δ is smaller than B, store the value of the larger of (f_N,M+ΔD−T) and 0 and (BD+MTU) in the register Q as the new queue length,
- when this Δ is not smaller than B, take (q−T) as the new queue length and store it in the register Q.

Example 2

In order to perform the processing of the flowcharts above at high speed, it is preferably implemented in semiconductor circuits (hardware). However, in order to implement it in hardware, the processing procedure must be described in hardware description language (HDL) and converted to circuitry, but the function (max) that finds the maximum value is used in Table 6 above, yet this operate cannot be used in the current state. In addition, division is used in the above processing procedure and this may complicate the circuitry, or cause the processing to slow down, among other problems. These problems are solved by setting T=2^pD (where p is an integer).

In this case, the first-stage processors P_1nperform processing according to the flowchart shown in FIG. 21. If this processing were written in the form of a program, it would be as follows.

	TABLE 8

	First Stage (P_1n: 1≦n≦8)
	l_n, l_n−1	: packet length
	t_n, t_n−1	: arrival gap time
	g_n−1	: temporary variable for queue length
	f_n,1, t_n,1	: output
	T	: period
	D=2^k	: unit length of the delay element
	B	: number of the delay elements
	for each processor n, in parallel (n = 1 to N)
	begin
	if (l_n−1= 0) t_n,1= t_n;
	else t_n,1= t_n−1;
	g_n−1= t_n−1⁽²⁾+ l_n−1;
	if (l_n= 0) f_n,1= g_n−1;
	else begin
	if (g_n−1⁽²⁾≦ t_n⁽²⁾) Δ_n= g_n−1⁽¹⁾;
	else Δ_n= g_n−1⁽¹⁾+ 1;
	f_n,1= t_n⁽²⁾+ l_n+ 2^kΔ_n;
	end;
	end

In Table 8 above, taking N to be the number of lightpaths, n to be the lightpath number, and the period T to be set as a power of 2 times an appropriately set time unit (τ), or namely T=2^aτ (where a is a nonnegative integer), while T is to be no greater than the minimum packet length represented as the length of an integral multiple of the time unit (τ), the above reading means performs reading at each period T, and for the optical packets of the n^thlightpath that pass through this reading means at a specified time, the length of the optical packet is l_nrepresented as a length which is an integral multiple of the time unit, the pre-period is expressed as an integral multiple of the time unit that is no greater than the pre-period, and for n=0, l_n=t_n=0, and the delay of the plurality of delay elements, in a constitution where the delays can be arranged such that they increase by the delay time D, D is given as 2^k(where k is a nonnegative integer), and, as the method of representation, the integer corresponding to the least significant bit k of the binary representation of the integer x is given as x⁽²⁾, and the binary number consisting of the remaining most significant bits after removing the aforementioned least significant bit k is given as x⁽¹⁾, this processor is provided with N lightpaths on the input side, and performs the following processing on the integers n from 1 to N

1) When l_n−1=0 is true, output t_n, as t_n,1, or
- when l_n−1=0 is not true, output t_n−1as t_n,1,
2) in addition, taking g_n,1=l_n−1+t_n−1,
3) when l_n=0 is true, output g_n−1as f_n,1, or
- when l_n=0 is not true, when g_n−1is less than t_n, set g_n−1⁽¹⁾to Δ_n,
- and when g_n−1is not less than t_n, set (g_n−1⁽¹⁾+1) to Δ_n, and
4) set (t_n⁽²⁾+l_n+Δ_nD) to the output f_n,1.

The processing procedures for the second stage and third stage are shown in FIGS. 22 and 23, respectively. If this processing were written in the form of a program, it would be as shown in Tables 8 and 9, respectively.

	TABLE 9

	Second Stage (P_2n: 3≦n≦8)
	fn,1 , fn−2,1	: last position of the combined packet
	t_n,1, t_n−2,1	: arrival gap time
	Δ_n	: temporary variable
	f_n,2, t_n,2	: output
	for each processor n, in parallel (n = 3 to N)
	begin
	if (f_n−2,1= 0) t_n,2= t_n,1;
	else t_n,2= t_n−2,1;
	if f_n,1= 0 f_n,2= f_n−2,1;
	else begin
	if (f_n−2,1⁽²⁾≦ t_n,1⁽²⁾) Δ_n= f_n−2,1⁽¹⁾;
	else Δ_n= f_n−2,1⁽¹⁾+ 1;
	f_n,2= f^n,1+ 2^kΔ_n;
	end;
	end

	TABLE 10

	Third Stage (P_3n: 5≦n≦8)
	f_{n,2 ,}f_n−4,2	: last position of the combined packet
	t_{n,2 ,}t_n−4,2	: arrival gap time
	Δ_n	: temporary variable
	f_{n,3 ,}t_n,3	: output
	for each processor n, in parallel (n = 5 to N)
	begin
	if (f_n−4,2= 0) t_n,3= t_n,2;
	else t_n,3= t_n−4,2;
	if f_n,2= 0 f_n,3= f_n−4,2;
	else begin
	if (f_n−4,2⁽²⁾≦ t_n,2⁽²⁾) Δ_n= f_n−4,2⁽¹⁾;
	else Δ_n= f_n−4,2⁽¹⁾+ 1;
	f_n,3= f^n,2+ 2^kΔ_n;
	end;
	end

These procedures are generally performed as follows. Generally, in the buffering method of a device that is provided with the m^th-stage (n−β)^thprocessor P(m,n) that, for the integers n from β+1 to N where β=2^m−1, calculates the content of the n^thregister R(m−1,n) connected to the previous-stage processors P(m−1,n) and the content of the (n−β)^thregister R(m−1,n−β) and gives the output of (f_n,m, t_n,m).

1) When f_n−β,m−1=0 is true, output t_n,m−1as t_n,m, or
- when f_n−β,m−1=0 is not true, output t_{n−β, m−1}as t_n,m,
2) in addition, when f_n,m−1=0 is true, output f_n−β,m−1as f_n,m, or
- when f_n,m−1=0 is not true,
2-1) when f_n−β,m−1⁽²⁾is less than or equal to t_n,m−1⁽²⁾, set f_n−β,m−1⁽¹⁾to Δ_n,
2-2) when f_n−β,m−1⁽²⁾is not less than or equal to t_n,m−1⁽²⁾, set (f_n−β,m−1⁽¹⁾+1) to Δ_n,
3) set (f_n,m−1+2^kΔ_n) to the output f_n,m.

At the stage of providing output of Δ_n, the processing according to the flowchart shown in FIG. 24 is performed. If this were written in the form of a program, it would be as follows.

	TABLE 11

	(m+1)_m−1-th Stage ,Delay time, (P_M+1,n: 1≦n≦N) : M=log2 N
	f_n−1,M_m−1	: last position of the combined packet
	l_n	: packet length arrived M periods before
	t_n	: arrival gap time M periods before
	t_n-1.M	: arrival gap time
	q	: queue length
	Δ_n	: delay
	for each processor n, in parallel (n = 1 to N)
	begin
	if (l_n= 0) exit;
	if (f_n−1,M= 0)
	begin
	if (q⁽²⁾≦ t_n⁽²⁾) Δ_n= q⁽¹⁾;
	else Δ_n= q⁽¹⁾+ 1;
	end
	else begin
	if (q⁽²⁾≦ t_n−1,M⁽²⁾) Δ₁= q⁽¹⁾;
	else Δ₁= q⁽¹⁾+ 1;
	if (f_n−1,M⁽²⁾≦ t_n⁽²⁾) Δ_n= Δ₁+ f_n−1,M⁽¹⁾;
	else Δ_n= Δ₁+ 1 + f_n−1,M⁽¹⁾;
	end ;
	if (Δ_n< B) Packet n is given delay Δ_n;
	else Packet n is discarded;
	end

Table 11 presents the following processing for the integers n from 1 to N when the length of the queue is q.

Assuming that M=log₂N,

1) when l_n≠0 is not true, end processing,

2) when l_n≠0 is true,

2-1) when f_n−1,M=0 is true and,

- when q⁽²⁾is less than or equal to t_n⁽²⁾, set q⁽¹⁾to Δ_n,
- when q⁽²⁾is not less than or equal to t_n⁽²⁾, set (q⁽¹⁾+1) to Δ_n,
  2-2) when f_n−1,M=0 is not true and,
- when q⁽²⁾is less than or equal to t_n−1,M⁽²⁾, set q⁽¹⁾to Δ₁,
- when q⁽²⁾is not less than or equal to t_n−1,m⁽²⁾, set (q⁽¹⁾+1) to Δ₁, and moreover,
  2-3-1) when f_n−1,Mis less than or equal to t_n⁽²⁾, set (Δ₁+f_n−1,M⁽¹⁾) to Δ_n,
  2-3-2) when f_n−1,Mis not less than or equal to t_n⁽²⁾, set (Δ₁+1+f_n−1,M⁽¹⁾) to Δ_n,
  3) output Δ_nwhen Δ_nis less than B,
- but if not less, discard the corresponding optical packet.

In addition, at the stage of updating the queue length, processing is performed according to the flowchart shown in FIG. 25. If this were written in the form of a program, it would be as follows.

	TABLE 12

	(m+1)-th Stage, Queue length, : M=log2 _N
	f_N,M	: last position of the combined packet
	t_N,M	: arrival gap time
	q	: queue length
	Δ	: temporary variable
	q_new	: output queue length
	if (f_N,3= 0)
	begin
	if (q > T) q_new= q − T;
	else q_new= 0;
	end
	else begin
	if (q⁽²⁾≦ t_N,M⁽²⁾) Δ = q⁽¹⁾;
	else Δ = q⁽¹⁾+ 1;
	if (Δ < B)
	begin
	if (f_N,M+ 2^kΔ > 2^kB + MTU)
	q_new= 2^kB + MTU − T; (*)
	elsif (f_N,M+ 2^kΔ > T) q_new= f_N,M+ 2^kΔ − T;
	else q_new= 0;
	end
	else q_new= q − T;
	end

In Table 12 above, the processing is performed by the following procedure. As described above, the MTU (maximum transfer unit) is generally set.

When M=log₂N,

1) when f_N,M=0 is true,
1-1) when q is greater than T, (q−T) is set as the new queue length (q_now),
1-2) when q is not greater than T, q_new=0 is set,
2) when f_N,M=0 is not true,
2-1) when q⁽²⁾is less than or equal to t_N,M^{(2)l, q}⁽¹⁾is set as Δ,
- when q⁽²⁾is not less than or equal to t_N,M⁽²⁾, (q⁽¹⁾+1) is set as Δ,
3-1) when this Δ is smaller than B and,
3-1-1) when (f_N,M+2^kΔ) is greater than (2^kB+MTU), (2^kB+MTU−T) is set as q_new,
3-1-2) when (f_N,M+2^kΔ) is not greater than (2^kB+MTU) and,
3-1-2-1) when f_N,M+2^kΔ is greater than T, (f_N,M+2^kΔ−T) is set as q_new,
3-1-2-2) when f_N,M+2^kΔ is not greater than T, q_now=0 is set,
3-2) when this Δ is not smaller than B, (q−T) is set as q_new.

In each period, if the buffer is already overflowing with optical packets at the time that the queue length is updated, perform the process of discarding newly arriving packets in the same manner as in the conventional round-robin scheme. If not, perform the queue length update assuming that all arriving optical packets can be added to the queue. The computations for appropriately storing all arriving optical packets are to be performed with a plurality of processors.

Example 3

When the above buffering device in the optical packet switch for asynchronous variable-length optical packets is used for buffering in an optical packet switch for synchronous fixed-length optical packets, the parallel processing of FIG. 26 can be simplified to give the constitution of FIG. 27. Similarly, in the same manner as the constitution illustrated in FIG. 13, with the constitution using the dotted-line portion that sequentially sends data forward, the variables can be set to those of synchronous fixed-length to give the constitution of FIG. 28. In FIGS. 27 and 28, the number of variables used in processing is reduced. The number of variables can be reduced because all of the optical packet lengths become identical, so it is possible to use the number of optical packets instead of the optical packet length in the above process. In addition, the respective optical packets are synchronized, so the processing of the pre-period of optical packets can be omitted. The processing to this end is performed as follows.

A device that accepts input of synchronous fixed-length optical packets from a plurality of lightpaths, gives each of the optical packets one of B different delay times and outputs same to a single optical path, comprising:

reading means that reads the presence of optical packets on the respective lightpaths,

a queue consisting of a plurality of delay elements that temporarily store the plurality of optical packets and give each optical packet one of B different delay times, an optical switch that directs the optical packet to one of the plurality of delay elements, an optical demultiplexer connected by lightpaths to the plurality of delay elements, and an output lightpath connected by a lightpath to the demultiplexer,

and in order to prevent the optical packets from overlapping in the output lightpath, a controller that uses the delay times of the delay elements and the presence information for optical packets thus read to determine by computation the delay element used for temporary storage, and

this controller is such that:

when the plurality of lightpaths are given sequential numbers n from 1 to N, and in the following, cases in which there is none corresponding to n can be neglected,

for the number of optical packets in the respective lightpaths,

- as the arrangement of the first stage of processors and registers, processors P(1,n) that perform the calculation of the number of optical packets on the (n−1)^thlightpath and the optical packet numbers of optical packets on the n^thlight path, and registers R(1,n) that store the results of calculation are disposed,

as the arrangement of the m^th(2≦m≦log₂A) stage, processors P(m,n) that perform the calculation of the registers R(m−1,n−2^m−1) or copies thereof and the registers R(m−1,n) or copies thereof, and registers R(1,n) that store the results of calculation are provided,

for the integers M where 2^M=N, on the (M+1)^thstage are provided a processor and a register Q that stores the information at the tail end of the queue length,

moreover, the delay applied to the first synchronous fixed-length optical packet is determined from the optical packet presence information for the first lightpath or a copy thereof and the value of the register Q, and

for values of n greater than 1, processors P(M+1,n) are used to determine the delay applied to the n^thsynchronous fixed-length optical n^thpacket, by calculation on the optical packet presence information for the nth lightpath or a copy thereof and the values of the register R(M,n−1) and the register Q, thereby coordinating the delay times given to the synchronous fixed-length optical packets of the respective lightpaths.

We shall first describe the first-stage processors P_1n(P_1n: 2≦n≦N). Here, for the N processors P_1n(where 2≦n≦N), the following processing is performed after the time of (log₂N+1)T until after the time (log₂N+2). Here, (log₂N+1) is the number of steps in which the following process is performed as parallel processing and T is the time per 1 step. This process must not end before the optical packets reach the optical switch for directing them to the respective delay elements.

In the case of the first stage, the input to the variables l_n−1and l_nthat indicate the presence of optical packets arriving at port (n−1) and port n is 1 when a packet has arrived and zero when no packet has arrived. The first-stage processors perform the following processing based on this information, and provides output of the presence information f_nfor optical packets arriving at the two ports.

Start

l₀=0

for each processor n, in parallel (n:=1 to N)

f_n,1:=l_n−1+l_n;

End

In addition, the second stage (P_2n: 3≦n≦N) is processed as follows.

For the (N−2)^thprocessor P_2a(where 3≦n≦N), the optical packet presence information obtained from the first-stage processors is input as:

- f_n,1which represents the number of optical packets arriving at port (n−1) through port n,
- f_n−2,1which represents the number of optical packets arriving at port max((n−3),1) through port (n−2).

In addition, the following process is used to output f_n,2, which is the number of optical packets arriving at port max(n−3,1) through port n.

Start

for each processor n, in parallel (n:=3 to N)

f_n,2:=f_n−1+f_n−2,1;

End

In addition, the third stage (P_n: 5≦n≦N) is processed as follows.

For the (N−4)^thprocessor P_0n(where 5≦n≦), the optical packet presence information obtained from the second-stage processors is input as:

- f_n,2which represents the number of optical packets arriving at port (n−3) through port n,
- f_n−4,2which represents the number of optical packets arriving at port max((n−7),1) through port (n−4).

In addition, the following process is used to find and output f_n,3, which is the number of optical packets arriving at port max(n−7,1) through port n.

Start

for each processor n, in parallel (n:=5 to N)

f_n,3:=f_n−2+f_n−4,2;

End

In general, the m^thstage (P_mn: 2^m−1+1≦n≦N) is processed as follows.

Taking β=2^m−1, for the (N−2^m−1)^thprocessor P_mn(where β+1≦n≦N), the optical packet presence information, obtained from the (m−1)^th-stage processors is input as:

- f_n,m−1which represents the number of optical packets arriving at port (n−2^m−1+1) through port n,
- f_n−β,m−1which represents the number of optical packets arriving at port max((n−2^m−1+1),1) through port (n−2^m−1).

In addition, the following process is used to find and output f_n,m, which is the number of optical packets arriving at port max((n−2^m−1+1),1) through port n.

Start

for each processor n, in parallel (n:=2^m−1+1 to N)

f_n,m:f_n−m−1+f_n−β,m−1;

End

With the above processing, it is possible to use pipeline processing in a parallel configuration of m stages of processors to find the number of optical packets arriving at port 1 through port n, for n=1, 2, . . . , N.

In addition the delay is found for the (M+1)^thstage. FIG. 29 presents a flowchart for this purpose. In addition, an example of a program corresponding to this flowchart is presented below.

For the processors P_M+1,n(where 1≦n≦N), upon taking input of f_n−1,Mas the number of optical packets arriving from ports 1 to n−1, l_nwhich indicates the presence of an optical packet arriving at port n, and the queue length q, the value of the delay Δ_nis output.


	Start
	for each processor n, in parallel (n:=1 to N)
	begin
	if (l_n=0) exit;
	Δ_n:=q+f_n−1,M;
	if (Δ_n<B) Packet n is given delay Δ_n;
	else Packet n is discarded;
	end
	End

At the same time as the derivation of the delay, separate processors P_qperform an update of the stored buffer occupancy.

The input is f_N,M, the number of optical packets arriving at all ports.

The internal variables are Δ and the queue length q.

The output is the new queue length q_new.

q_new:=max((min(f_N,M+q,β)−1,0);

When the above process is performed by the conventional round-robin type method, the following can be done.


	Start
	for n:=1 to N do
	begin
	if (l_n=0)
	begin
	if f<B
	begin
	Packet n is given delay f× T,
	f.=f+ 1;
	end
	else Packet n is discarded;
	end
	end
	f.=max(f−1,0);
	End

When round-robin type processing is performed in this manner by the conventional method, the loop must be traversed N times, so the processing time from start to end per processor becomes a much longer time than in the case of parallel processing according to the present invention as above. In other words, if the processing time of a processor according to the present invention is taken to be the time unit, the round-robin type processing requires N time units to process one packet per port.

The optical packets transmitted over various lightpaths are typically subject to shifting out of synchronization, but by using a previously reported optical synchronization device, it is possible to obtain synchronized optical packets. In addition, by setting the optical packet length of the largest transmitted optical packet to the respective optical packet length, they can be made fixed length. Thus, synchronized fixed-length optical packets can be used by this method.

In addition, the above method of preparing synchronized fixed-length optical packets can also be applied even to the case of asynchronous variable-length optical packets, so it is clear that after converting asynchronous variable-length optical packets to synchronous fixed-length packets, the buffering device can be applied to these synchronous fixed-length packets and thus the asynchronous variable-length optical packets can be buffered so that they do not overlap on output.

Table 13 presents the results of the method according to the present invention in comparison to the results when dendriform delay line buffer units are disposed and the method recited in Document 6 is used with a buffer managers placed in each, and the method recited in Document 5 for finding the delay by parallel processing, for each of the cases of N=8, 64, 256 and 1024.

TABLE 13

Number of processors, Np. processing delay, TD. and processing time, Tp. per processor when input N = 2k.
	con-
	ventional	document 6	document 5	Present Invention
N	k	N_P	T_D	N_P	T_D	T_P	N_P	T_D	T_P	N_P	T_D	T_P

		1	O(N)	O(N^1/2)	O(N^1/2)	O(N^1/2)	O(N)	O(log₂N)		O(N log₂N)	O(log₂N)	O(1)
		1	N	N^1/2+ 1	2N^1/2	N^1/2	N	log₂N + 1	log₂N + 1	N log₂N + 2	log₂N + 1	1
16	4	1	16	5	8	4	16	5	5	66	5	1
64	6	1	64	9	16	8	64	7	7	386	7	1
256	8	1	256	17	32	16	256	9	9	2,050	9	1
1,024	10	1	1,024	33	64	32	1,024	11	11	10,242	11	1

Here, N=2^kis the number of inputs, N_p, T_Dand T_Pare the number of processors, processing delay and processing time per processor, respectively. As is evident from this Table 13, while the number of processors used is large in the present invention, the processing delay (T_D) and processing time per processor (T_P) can be made small. In addition, the characteristic of the present invention in that the processing time per processor does not change even if N (or k) that is the number of inputs increases is further superior in comparison to Document 5 above. In this manner, the characteristic of being able to reduce the processing delay (T_D) and processing time per processor is necessary in order to perform high-speed buffering. In particular, the latter characteristic of being able to reduce the processing delay per processor allows more processes to be performed per time unit. For example, when N=1024, where only one process per input could be performed during 111 time units according to Document 5 above, with the present invention, 11 processes can be performed per input during 11 time units. In this manner, one can see that it is possible to avoid rapid increases in the maximum processing time per processor even when the number of inputs increases. This meritorious effect is an object of the present invention, and the effect thereof becomes even more prominent as N increases further,

Example 4

In the followings, we describe two prioritized buffer management methods in parallel and pipeline processing architecture. At first, we directly apply PBS (partial buffer sharing) to the multiprocessing architecture. We then point out its problem and design a new prioritized buffer management method, called PBFS (partial buffer fair sharing), which is an extended version of the PBS. PBFS is closer to the multiprocessing architecture rather than PBS from the viewpoint of buffer utilization. We describe the processing mechanism in the multiprocessing architecture.

In PBS method, threshold TH is predetermined. When the queue is shorter than the threshold (q<TH), every packet is allowed to enter the queue. When the queue length is equal to or longer than the threshold and is shorter than the buffer size (TH≦q<B), only high-priority class (class-1) packets are allowed to enter the queue while low-priority class (class-2) packets are discarded. The mechanism can be combined into the round robin scheduling, by replacing condition “if q<B” in the forth line in the following procedure;


	0. Procedure round-robin
	1. For n:=1 to N do
	2. Begin
	3. If (l_n=1) then begin
	4. If q < B then begin
	5. Packet n is given delay qD;
	6. q:=q+1; End
	7. Else Packet n is discarded;
	8. End
	9. End
	10. q:= max(q−1, 0) ,

into condition “if ((q<B) and (c_n=HIGH) or (q<TH) and (c_n=LOW))”, as follows;


0. Procedure PBS
1. For n:=1 to N do
2. Begin
3. If (l_n=1) then begin
4. if ((q<B) and (c_n= HIGH) or (q<TH) and (c_n= LOW)) then begin
5. Packet n is given delay qD;
6. q:=q+1; End
7. Else Packet n is discarded;
8. End
9. End
10. q:= max(q−1, 0); ,

where c_nindicates the priority class of the packet arriving at port n.

Parallel PBS is a direct application of PBS. In the non priority case in Example 3, the number of arriving packets is calculated in the front log₂N pipeline stages to be used for the delay calculation and the update of the queue length. Since PBS does not allow to enter class-2 packets when the queue reaches the threshold, calculating the number only of packets including class-2 packets results in overestimation of the queue length. We therefore introduce another prefix-sum, in which arrivals of class-1 packets are used as elements. In each processor in the pipeline stages, the number of class-1 packets as well as the number of packets (namely, sum of the number of class-1 packets and the number of class-2 packets) are calculated.

We first describe procedures in front log₂N pipeline stages for parallel prefix operation. The first-stage processors perform the first operation of the parallel prefix. For processor P_1,n(2≦n≦N), values l_nand l_n−1are inputted. Values c_nand c_n−1respectively indicating the priority class of the packet at ports n and n−1, are also inputted. The value is 1, if the packet belongs to class-1, and it is 0 otherwise. Processor P_1,nperforms the procedure in the following procedure to generate f_n,1and g_n,1, which respectively represent the number of packets and the number of class-1 packets at the two ports.


	0.	Procedure first stage of parallel PBS;
	1.	For each processor P_l,n, in parallel (n:= 2 to N)
	2.	Begin
	3.	f_{n, l}:= l_n+ l_n·1;
	4.	g_{n, l}:= c_nl_{n + c}_n·1l_n·1;
	5.	End,

The kth-stage processors are devoted to the kth operation of the parallel prefix. For processor P_k,n(2≦k≦log₂N, 2^k−1+1≦n≦N), the inputs are connected to the registers of the two (k−1)th-stage processors P_k−1,nand P_k−1,2^k−1. Processor P_k,nthus receives f_n,k−1, g_n,k−1, f_n−2_k−1_,k−1, and g_n−2_k−1_,k−1; it then performs the procedure in the following procedure to generate f_n,kand g_n,k, which respectively represent the number of packets and the number of class-1 packets at ports max(n−2^k+1, 1) through n.


	0.	Procedure k-th stage of Parallel-PBS;
	1.	For each processor P_k,n, in parallel (n:= 2^k·1+ 1 + 1 to N)
	2.	Begin
	3.	f_n,k·1:= f_n,k·1+ f_n− 2^k·1_,k·1;
	4.	g_n,k·1;= g_n,k·1+ g_n− 2^k·1_,k·1;
	5.	End,

By performing the above procedures at the front log₂N stages, processors in the (log₂N)th stage generate a set of prefix sums. The prefix sums are the number of packets and the number of class-1 packets both arriving at ports 1 through n(n=1, . . . , N).

N processors at the (log₂N+1)th stage are devoted to delay determination. For processor P_{(log2 N+1),n}(1≦n≦N), the inputs are connected to the register of the (log₂N)th stage processor, P_{(log2 N+1),n−1}. Processor P_{(log2 N+1),n}thus receives f_{n−1,log2 N}and g_{n−1,log2 N}. At the same time, the processor also receives the queue length at the beginning of the cycle (q_start) and the original information about packet arrival (l_n) and its priority class (c_n). The delay to a packet at port n is determined by sum of q_startand the number of packets, or, the number of class-1 packets. Since PBS allows to enter class-2 packets only when the queue is smaller than the threshold, delay determination should depend on the current status of the queue length. Namely,

When the queue length is equal to or greater than the threshold at the beginning of the cycle (q_start≧TH), the queue length during the cycle is also equal to or greater than the threshold. Class-2 packets are not allowed to enter the queue. The delay to a class-1 packet is determined as the sum of q_startand the number of class-1 packets, Δ_n=q_start+g_{n−1,log2 N}.

When the queue is smaller than the threshold at the beginning (q_start<TH), the queue length may reach the threshold. However, Parallel PBS gives a packet the delay that is the sum of the queue length at the beginning of the cycle and the number of packets, Δ_n=q_start+f_{n−1,log2 N}, regardless of the priority class of the packets.

If we strictly apply the PBS, other variables are needed in addition to f_*,*'s and g_*,*'s and delay determination becomes very complicated. The complication makes the speed of parallel and pipeline processing lose. We therefore make the delay determination simple. Finally, the class-1 packet at port n is given delay Δ_nif Δ_n<B, and is discarded otherwise. The class-2 packet at port n is given delay Δ_nif Δ_n<TH, and is discarded otherwise. The following procedure shows the pseudo-code for processor P_{(log2 N+1),n}.


0.	Procedure of (log₂N + 1)th stage of Parallel PBS
1.	For each processor P_{(log2 N+1),n}in parallel (n:= 1 to N)
2.	Begin
3.	If (l_n= 1) then begin
4.	If (q_start< TH) then Δ_n= f_{n·1 ,log2 N}
5.	else Δ_n= q_start+ g_{n·1 ,log2 N}
6.	if ((Δ_n< B) and (c_n= HIGH) or (Δ_n< TH) and (c_n= LOW))
7.	then Packet n is given delay Δ_nD;
8.	Else Packet n is discarded;
9.	End
10.	End,

At the last stage, the queue length is also updated at processor P_q. The input of the processor is connected to the register of processor P_{log2 N,n}. Processor P_qthus receives f_{n,log2 N}and g_{n,log2 N}. If queue is smaller than the threshold at the beginning of the cycle, the queue length is updated as follows;

q:=max(min(q_start+f_{N,log2 N}, B)−1, 0)

Otherwise that is, if the queue is equal to or larger than the threshold, the queue length is updated as follows.

- q:=max(min(q_start+g_{N,log2 N}, B)−1, 0)
  By using the above update procedure, the overestimation of the queue length is avoided.

Parallel PBS can increase throughput but the buffer is not utilized well. For example, consider a queue of which length is two (q_start=2), where one packet has been served and another is waiting. Assume that four packets (classes 1, 1, 2, and 1 in the scheduled order) designated to the same output port arrive at the packet switch in this cycle. Threshold TH is three. In this case, after the first class-1 packet is allowed to enter the queue and is given delay two, the queue reaches the threshold. The second class-1 packet is given delay three. The class-2 packet is discarded. Nevertheless, the last class-1 packet is given delay five because the delay is determined by means of the number of arriving packets. As a result, we have free space between the second class-1 packet and the last one.

We extend PBS in order to avoid the free space. The condition for packet buffering (the forth line in the Procedure round-robin) is replaced by “if ((q<B) and (c_n=HIGH) or (q_start<TH) and (q<B) and (c_n=LOW))”. Namely, if queue length at the beginning of the cycle is smaller than the threshold, class-2 packets are allowed to enter the buffer. Since packets in the same cycle are fairly treated except for buffer overflow, we call this method partial buffer fair sharing (PBFS).

PBFS can also be introduced in the multiprocessing architecture. In the Parallel PBFS, the procedures in the front log₂N stages and procedure in processor P_qare identical to those in the Parallel PBS. Delay determination procedures in the last stage are different. The delay to a packet is temporally determined by the same method as that of Parallel PBS. The difference exists in the decision of packet buffering; the class 2 packet at port n is given delay Δ_nif q_start<TH and if Δ_n<B, and is discarded otherwise. The following procedure shows the pseudo code for processor P_{(log2 N+1),n}.


0. Procedure of (log₂N + 1)th stage of Parallel PBFS
1. For each processor P_{(log2 N+1),n}in parallel (n:=1 to N)
2. Begin
3. If (l_n=1) then begin
4. if (q_start< TH) then Δ_n:= q_start+ f_n−1,log2N
5. else Δ_n= q_start+ g_{n−1 ,log2 N}
6. if ((Δ_n<B) and (c_n= HIGH) or (q_start<TH) and (Δ_n<B) and (c_n=
LOW))
7. then Packet n is given delay Δ_nD;
8. Else Packet n is discarded;
9. End
10. End,

Parallel PBFS can use buffer more efficiently than Parallel PBS does because PBFS does not generate the free space, which is generated by Parallel PBS.

In order to demonstrate the hardware scale and the feasibility of high-speed management, we conducted a simulation of the buffer management hardware with partial buffer fair sharing (PBFS) priority control based on a multi-processor configuration. We simulated our configuration after performing a place-and-route operation on a 0.13-μm FPGA (Field Programmable Gate Array) wiring process, in which 79,040 logic elements are integrated and 830 I/O pins are prepared. The specifications of this configuration circuit are given in the table below.

TABLE 14

Specifications


	Clock speed (f_max)	156.4	MHz
	Port speed (C)	80.0	Gpbs
	Number of inputs (N)	8
	Packet length (L)	64	bytes
	Number of FDLs (B)	15
	Unit length of FDLs (D)	1.56	m (64 bytes)

As indicated by the table above, this configuration circuit has a clock speed of 156.4 MHz. The circuit provides priority control for 8-input optical packet switch. The number of logic elements required to implement this circuit was 456. Note that from the equation below, the circuit speed at this clock speed becomes 80 Gbps.
C=8L×f_max
Here, C, L and f_maxare the channel speed (Gbps), packet length (bytes) and the clock speed (MHz), respectively. The breakdown of the critical path in this imaging circuit was 44% cell delay and 56% wiring delay.

The maximum number of input ports can be estimated by using configuration-circuits with a smaller number of input ports. The upper limit for input ports is limited by the number of I/O pins and number of FPGA logic elements that can be used. In the configuration circuit, if N is the number of input ports, then (6N+19) X/O pins are required. The number of ports that can be obtained therefrom is 128.

On the other hand, FIG. 30 presents an estimate of the limit from the number of logic elements.

The scale of a multiprocessor circuit is on the order of Nlog₂N. In addition, the number of logic elements used in the reference function is 32Nlog₂N. From FIG. 30, one can see that the value of the reference function at N=128 is smaller than the number of FPGA logic elements prepared, so it is possible to construct an configuration circuit expanded to 128 ports. While the channel (port) speed typically decreases as the number of ports increases, in the case of 128 ports, the channel speed is greater than half that in the case of 8 ports described above. Even when considering this decrease in channel speed, the overall throughput of packet switch, which is derived from product of the number of input ports and the channel speed per port, reaches 5120 Gbps.

INVENTORS:

Harai, Hiroaki

THIS PATENT IS REFERENCED BY THESE PATENTS:

Patent	Priority	Assignee	Title
8306047,	Jul 29 2009	University of Southern California	Packet switch with separate look ahead, computation, and shift phases
9961420,	Feb 27 2014	National Institute of Information and Communications Technology	Optical delay line and electronic buffer merged-type optical packet buffer control device

THIS PATENT REFERENCES THESE PATENTS:

Patent	Priority	Assignee	Title
7106967,	Sep 04 2001	F POSZAT HU, L L C	Optical packet switching apparatus and methods

ASSIGNMENT RECORDS Assignment records on the USPTO

Executed on	Assignor	Assignee	Conveyance	Frame	Reel	Doc
Jul 09 2004		National Institute of Information and Communications Technology Incorporated Administrative Agency	(assignment on the face of the patent)
Aug 09 2004	HARAI, HIROAKI	National Institute of Information and Communications Technology Incorporated Administrative Agency	ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS	015919	0520	pdf

MAINTENANCE FEES AND DATES: Maintenance records on the USPTO

Date	Maintenance Fee Events
Jun 17 2013	ASPN: Payor Number Assigned.
Sep 27 2013	M1551: Payment of Maintenance Fee, 4th Year, Large Entity.
Sep 26 2017	M1552: Payment of Maintenance Fee, 8th Year, Large Entity.
Nov 22 2021	REM: Maintenance Fee Reminder Mailed.
May 10 2022	EXP: Patent Expired for Failure to Pay Maintenance Fees.

Date	Maintenance Schedule
Apr 06 2013	4 years fee payment window open
Oct 06 2013	6 months grace period start (w surcharge)
Apr 06 2014	patent expiry (for year 4)
Apr 06 2016	2 years to revive unintentionally abandoned end. (for year 4)
Apr 06 2017	8 years fee payment window open
Oct 06 2017	6 months grace period start (w surcharge)
Apr 06 2018	patent expiry (for year 8)
Apr 06 2020	2 years to revive unintentionally abandoned end. (for year 8)
Apr 06 2021	12 years fee payment window open
Oct 06 2021	6 months grace period start (w surcharge)
Apr 06 2022	patent expiry (for year 12)
Apr 06 2024	2 years to revive unintentionally abandoned end. (for year 12)